AI News Archive: June 22, 2026 — Part 9
Sourced from 500+ daily AI sources, scored by relevance.
- Some Electricians Think Building Data Centers Is for Sellouts
Big Tech is throwing big money into data center buildouts. As national opposition to the facilities grows, some workers are beginning to question whether it’s worth it.
Score: 33🌐 MovesJun 22, 2026https://www.wired.com/story/data-center-buildout-electricians-selling-out/ - ProfPrep Brings Unlimited AI-Generated Exam Questions to Professional Licensing — No Candidate Runs Out of Material
ProfPrep Brings Unlimited AI-Generated Exam Questions to Professional Licensing — No Candidate Runs Out of Material azcentral.com and The Arizona Republic
- Emergent Behavior in Agentic Swarms
What building an agentic swarm taught me about control, delegation, and letting go One of the most interesting challenges in designing agents for complex, long-horizon tasks is the ability to maintain a sort of equilibrium between agency and criticism while setting the task goals and striving to achieve them. This is arguably the “art” that drives the quality of the agentic harness. Why is this an “art”? Because it necessitates balancing the opposing forces between advancing forward towards the task goals and processing the correcting feedback. If the scale tips toward advancement, there’s a bigger chance of deviation from the original goals and rules. Too much pulling on the reins and the advancement might stop, tripping the agent into endless loops and companies pay dearly in tokens for this imbalance. The real problem is that there are many cases where it seems this equilibrium is achieved but after numerous cycles, mostly in production where the environment is more diverse than the testing grounds, something along the way knocks the system out of balance. A good agentic system is one that naturally aspires to optimal internal stability across changing external conditions via self-regulation (AKA homeostasis). Fig. 1. Homeostasis between agency and correction. Diagram by author, renderedwith the help of Claude. Homeostasis in any system is achieved via processing a monitoring signal of the baseline behavior. The better the compatibility between the signal (detailed, objective, recent, relevant, clear etc.) and the processing of that signal inside the feedback loop, the more potential the system has to obtain homeostasis. While a single agent needs to self-regulate its inference towards its long horizon goal, when designing composite agentic systems, additional dimensions are added and need to be balanced. Homeostasis in complex agentic systems When you need to design a cluster of agents for a complex task, the ability to divide the task, plan the execution, govern the process, share the data between interdependent phases and validate the outcome can become genuinely challenging. There are several architectures of composite agentic systems. The most basic and common pattern is an orchestrator agent that governs subagents and delegates the work to them and they in turn report back. If you add communication between the subagents, you’ll get an orchestrator and a team of agents that can share findings, coordinate and optimize their working process. The most intriguing architecture to me though, is of an agentic swarm where the architecture is less coherent and when designed correctly, could manifest an emergent implicit behavior. This type of implementation is often not meant for deterministic processes and can be very interesting for tasks that require a high degree of “creativity”. Fig. 2. Composite agentic architectures. Diagram by author, rendered with the help of Claude. I decided to develop an agentic swarm platform to experience firsthand what it takes to create such a system. I started off with sketching what are the main traits of an emergent behavior system. Usually with these systems, the starting point is a set of simple localized rules that guide the interaction between the individual units. The capacity for an emergent behavior is often directly connected to how well these rules balance opposing properties like the one we’ve mentioned earlier about agency and correcting feedback. Governance and task delegation are often opposite forces too. The optimal governance denotes a zero delegation. If an agent is designed for a full governance, it will avoid dividing the task and delegating it because this means a loss of control. If it is designed towards full delegation, it will lose track of the task and inflate the iterations sometimes indefinitely. Another thing worth mentioning is that within complex systems the individual units that comprise it are seldom truly monolithic. Instead, these systems tend to produce an emergent hierarchy that serves the overall functionality. For example: in a beehive the queen emits pheromones that regulate the whole hive activity. It suppresses worker reproduction, inhibits new queen rearing, maintains worker cohesion and modulates the colony’s defensive behavior. In sexually reproducing organisms, development begins with a single fertilized cell called a zygote. The zygote is totipotent, meaning it has the potential to give rise to all cell types of the organism. Through successive rounds of cell division, it forms a solid ball of cells known as a morula. As development progresses, these cells gradually lose their totipotency and become increasingly specialized. As the morula continues to develop, it transitions into the blastocyst stage, where cells communicate through molecular signals. These signals establish positional information that guides cells toward different developmental fates and helps coordinate the formation of the organism’s body plan. Fig. 3. Embryonic cell division. Image by author, rendered with the help of ChatGPT. I decided to loosely model the platform after the cellular division process of a growing organism because this is an emergent behavior that can perform an extremely complex task. The idea was to first start with a stem cell that will be able to create its own specialized type of cells and actively spawn them to achieve the task. Once this will display a dynamic enough expansion, phase two would be to produce a multi stem hierarchy where each one will grow its own sub task-oriented environment. The Tension Between Governance and Delegation The first hurdle I encountered was that LLMs agency naturally tends to self-complete a task. It will not naturally spawn and delegate tasks even if instructed so. LLMs are trained to strive for task completion. In our governance vs. delegation tension, they are leaning heavily toward governance. So, for the most part, phase one was failing to produce rich enough architecture. Even when I pushed hard via the system prompt to divide and delegate the task, this division was minimal and superficial just to comply with the system prompt directive. There have been cases where the main loop did the same work in parallel to the child cells to whom it supposedly delegated the work. That leads to the first principle of designing a swarm emergent behavior: Diversity often comes as an answer to constraints and limitations. As long as my stem cell agent had all the tools to perform the task, it preferred to do that all by itself. This was the easiest and most optimal way to achieve the goal. However, once a stem cell could only design new cells and spawn them but was deprived of web search tools for example, when it was prompted for research, it had to create new cells and delegate . So now the stem cell started to create new cells and delegate subtasks but there was never a real planning behind the task delegation. It was all spontaneous and directly driven by the attempt to work around the constraint. It came to mind that this was mainly from a linear line of work. When it had to invent and spawn new cells, it did so directly in response to the task at hand. Planning, unless specifically designated, is not really a tool that will be created spontaneously. The first principle came in handy here too. If you want a diversified behavior, you need to constrain. Choosing a predefined cell is limiting but it is a lesser effort than inventing a completely new cell. Adding to that the fact that predefined tools could have a description that will suggest when and for what it should be called (i.e. “use this cell when you need to plan for a complex task”) and suddenly planning does occur naturally when needed. So, I added predefined optional cell types on top of the ability to form and spawn a completely new cell type. The decision to keep the two options proved to be the right choice later on. Stepping Into Multi-Hierarchy It was time to move to phase two, from stem cell that spawns specialized cells to a stem cell that can spawn a multi hierarchy set of cells. I used my predefined cells to offer the initial stem cell (called zygote) a set of more specialized stem cells that will manifest specific workflows of processing data. For example: a tournament stem cell, rotating critique stem cell, cartographic stem cell etc. These in turn could spawn their own specialized cells. Keeping the option to invent new types of cells paid off at this point where specialized stem cells that faced a more specific task, could now easily fill in the functionality gap with their own versions of needed cells. When you scale up the number of agents, the next hurdle that becomes apparent is the imbalance between diversity and coordination. Reporting to the parent is not enough. This led to the second principle of agentic swarm systems: Effective agency is tightly coupled with the effectiveness of information sharing among the individual units. Cells need a medium to share information and coordinate. Asking the organism for simple tasks like forming a chain of five stem cells meant each cell needed to figure out its location in the chain and coordinate the spawning of additional cells until the final goal is met. This required communication skills far beyond a child completing a task and reporting results to its parent. Multiple messages needed to be passed throughout the chain. This was one of the most challenging tasks. Too much communication meant the agentic loop had to deal with excessive noise, and the work thread could easily be lost. I gave the platform a simple message bus. Potentially any cell can communicate with any other cell, but it needs to initiate a message. There were no mass public channels that will spread noise. Given the inherited linear processing of an LLM, almost all the communication happens between a parent cell and its children, keeping the information processing localized to a specific area, which is optimal in terms of agentic loop unnecessary noise. When cells started to communicate, the context window soon became a limitation. Context window management is a significant challenge across LLM systems broadly, with many different approaches and solutions. For this platform, I implemented a basic but effective solution for memory management in the form of key-value artifacts that cells could share with each other as a whiteboard, and added content compression when the context window exceeded a certain percentage of the max context window configuration. Designing for Implicit Control Now that swarm diversity has gone up and the cells could communicate effectively, governance that was nonissue on the single agent scale, became a huge problem on the overall swarm system scale. Once tasks had been delegated to subagents that in turn delegated them to additional cells, I could see that things are getting out of control. The formed organism could expand too much creating unnecessary and excessive work. Cells entered never ending processing loops that kept digging deeper into a task without any ability to know when and how to stop. So now I had a swarm, but I seldom got anything productive back from it. That led me to the third principle of designing an agentic swarm: Implicit control over a swarm is done over the design of a resource economy. In the single agent scale we’ve talked about homeostasis between critics and agency. This is an explicit tool that impact directly inside the agentic loop. Implicit control might sound as an oxymoron but looking at every complex system, this type of governance is done by regulating agency externally. The monetary system, for example, uses interest rates to regulate the flow of money and with that the whole economy. In biology, every action cost energy. Controlling the energy flow in the system would result in the type of governance I was looking for. I implemented an energy system where every cell is given an energy units budget from its parent. When it acts, it consumes the units and if running low, it can ask for additional energy, or it must end whatever it is doing and report back. The energy management was a game changer. It created a very flexible governance with which you could affect the size of the organism and the complexity of the task execution. Adding the option for different cost for every tool and cell type, I could implicitly favor certain tools and cells over the others. While playing with this option I noticed that if the energy flowed in only one direction, from parent to child, it created a very limiting effect on the possibility of structure developments. The system became very conservative in the way it handled processing and new spawning very similar to the “too much governance and no delegation” as it tried to preserve energy units. The flow had to go both ways. An organism spends energy on action, but it gains energy too when it consumes food. I figured that if a child cell has been able to complete its work, letting its parent harvest energy units from the outcome might encourage the parent to delegate more tasks and contradict its inherited behavior of self-dealing with the task. Furthermore, by tuning the rate of energy returned, the system could dynamically strike a homeostatic equilibrium. This indeed created a much better energy flow and resulted in a balance between the complexity and diversity of the system while maintaining the ability to receive highly practical products from the processing. What still surprise me in this project, sometimes to a point of frustration, it that engineering emergence is a different type of engineering. It’s similar to what it takes to be a good manager. You cannot be a good manager via forced control and micromanagement, you need to design the environment in which your organization grow. If you want to explore this space yourself, the platform is open-sourced on GitHub as zygote-oss (MIT licensed). It’s particularly well-suited for tasks where the solution space is genuinely open-ended and creativity matters more than determinism. Emergent Behavior in Agentic Swarms was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
Score: 32🌐 MovesJun 22, 2026https://pub.towardsai.net/emergent-behavior-in-agentic-swarms-88b2755bb1f0?source=rss----98111c9905da---4 - BrAInify launches its execution-focused AI learning platform in the UAE
brAInify is an AI-powered learning platform designed to bridge the gap between knowledge and execution to help in preparing the next generation of the UAE’s AI-enabled talent
- From support inbox to signal feed: Inside the AI workflow that won at Echelon Singapore 2026
A customer support inbox usually looks like a backlog: questions to answer, complaints to resolve, and product details to clarify before the next message arrives. For Aaryan Kandiah, it looked like something else: a live stream of business signals. That shift helped him win the AI Workflow Competition at Echelon Singapore 2026 with SignalDesk. A recent Nanyang Technological University graduate […] The post From support inbox to signal feed: Inside the AI workflow that won at Echelon Singapore 2026 appeared first on e27 .
- Anthropic's engineering leader says Claude Code is making programmers lonelier
Anthropic's engineering leader says Claude Code is making programmers lonelier Business Insider
Score: 32🌐 MovesJun 22, 2026https://www.businessinsider.com/anthropic-claude-code-ai-engineering-loneliness-fiona-fung - How do I build an AI medical scribe using speech-to-text?
Step‑by‑step guide to building an AI medical scribe with STT.
- WhatsApp gets new chief as Meta taps India’s CRED founder Kunal Shah and invests $900M in startup
WhatsApp gets a new boss, as Will Cathcart moves to a new role at Meta, while Shah steps down as CEO of Indian fintech giant CRED to replace Cathcart.
- Maximize project performance with an AI-powered construction ERP
How AI-native construction ERP platforms drive smarter decisions across projects and financials.
Score: 31🌐 MovesJun 22, 2026https://www.constructiondive.com/spons/maximize-project-performance-with-an-ai-powered-construction-erp/822757/ - From field notes to field intelligence: Greenda’s AI bet on everyday farming (Sponsored)
Munich-based AgriTech startup Greenda, founded in 2024, is building an AI-powered crop advisory platform aimed at helping small and medium-sized farms detect and manage pests faster, while giving cooperative technicians a clearer overview of what is happening across hundreds of plots. For many farmers, the problem is not a lack of knowledge. It is timing, […] The post From field notes to field intelligence: Greenda’s AI bet on everyday farming (Sponsored) appeared first on EU-Startups .
- TraPilot.ai Launches the World’s First AI-Native SEO Service Platform
TraPilot.ai Launches the World’s First AI-Native SEO Service Platform USA Today
- Apple’s productivity apps have a small but useful AI enhancement in macOS 27
While the new Siri AI is, of course, taking the headlines, there are a number of less spectacular but still useful AI enhancements in Apple’s upcoming operating systems. I recently highlighted the new AI photo editing tools , and Macworld has spotted a small but handy AI enhancement to Apple’s productivity apps …
Score: 30🌐 MovesJun 22, 2026https://9to5mac.com/2026/06/22/apples-productivity-apps-have-a-small-but-useful-ai-enhancement-in-macos-27/ - What if AI knew you better than you?
Aaru says it can predict our choices without asking us. Plus: lessons from the Asia Economic Summit and the realities of AI infrastructure in SEA.
- The Best AI Chatbots for 2026
The Best AI Chatbots for 2026 PCMag UK
- The art of AI integration: Growing your business with chatbots and human expertise
Singapore is working towards becoming a leader in artificial intelligence (AI), ranking among the top ten countries for AI startups and investments according to sources like Visual Capitalist and Techopedia. The rise of AI, particularly through AI chatbots, is making a big impact in Singapore and the world at large. These technologies are not just […] The post The art of AI integration: Growing your business with chatbots and human expertise appeared first on e27 .
Score: 30🌐 MovesJun 22, 2026https://e27.co/the-art-of-ai-integration-growing-your-business-with-chatbots-and-human-expertise-20240509/ - IDC Quanta: The Next Era of Tech Intelligence
IDC Quanta is an AI platform built on IDC’s proprietary research and intelligence, giving technology and strategy leaders on-demand access to the same trusted insights that have informed business decisions for decades. Instead of searching the open internet, users query a curated body of verified IDC research with every response citing the specific report or […] The post IDC Quanta: The Next Era of Tech Intelligence appeared first on IDC .
Score: 30🌐 MovesJun 22, 2026https://www.idc.com/resource-center/blog/idc-quanta-the-next-era-of-tech-intelligence/ - The Hard Part of AI
The Hard Part of AI Fortune
- How accurate is AI transcription for pharmaceutical drug names?
Analysis of AI transcription accuracy for drug nomenclature.
Score: 30🌐 MovesJun 22, 2026https://assemblyai.com/blog/ai-transcription-accuracy-pharmaceutical-drug-names - Rumble rebrands to RUM Group, launches AI business after Northern Data deal
The Longboat Key-based company is reorganizing into two businesses as it expands beyond its roots as a video-sharing platform.
Score: 30🌐 MovesJun 22, 2026https://www.bizjournals.com/tampabay/news/2026/06/22/rumble-rum-group.html?ana=brss_6150 - What is AI agent orchestration?
You start with one AI agent to save time. A month later, you've got prompts in a doc, outputs in Slack, half-finished automations in three places, and the same request getting handled a dozen different ways depending on who saw it first. That's what happens when businesses try to "do AI" by building roughly 43 agents with no plan in place to coordinate them. AI agent orchestration solves this problem. Instead of relying on a single, general-purpose AI agent to do everything (which rarely works),
- LLM-Driven Feature Discovery
We would often like to get a qualitative sense of a target model’s behaviors in important distributions (e.g. deployment, RL training, or evals). For example, we might want to discover novel behaviors , figure out what causes some target behavior to occur, or find surprising correlations between behaviors. In a recent short exploratory project, we tackled this problem via LLM-Driven Feature Discovery. Our method works as follows: Choose a dataset of model transcripts Split transcripts into three pieces: user turns, thoughts, and assistant responses. Ask a black box LLM autorater to generate a set of 10-20 “features” of each transcript piece. By feature we mean notable/interesting/important aspects of the transcript piece; we include the prompt we use below. Note that the autorater only sees one piece at a time. Get a semantic embedding for each generated feature Cluster the semantic embeddings separately for user, thoughts, and response features Ask a language model to name each cluster by giving it 100 random features for each cluster and asking it to “produce a single concise label (around 5 words) that captures the common theme of these features.”. During the project, we sometimes thought of this work as a sort of "black box SAE", since it was solving a similar problem as SAEs of featurizing model text, but without using model internals. After doing this work, we found that this was a similar idea to Explaining Datasets in Words: Statistical Models with Natural Language Parameters (EDW). EDW optimizes over directions in an embedding space and then maps those directions to natural language features (“predicates”). Thus, EDW’s output is similar to ours. However, our method is simpler in that it requires just one LLM call per prompt and does not require multiple steps of iteration. Additionally, our method is unsupervised; we don’t need a target to optimize the embedding directions against. EDW seems preferable if one aims to minimize the error of a specific statistical model with natural language features. Since this is preliminary work, we do not compare against EDW or other methods in the literature. We are not currently planning on pursuing this idea further, but would be interested if other members in the community expanded on it. A short summary of our main results: We focus our analysis on a dataset of 100k chat transcripts, for which we generate 20k user, thought, and response features. We find that: Many clusters describe interesting Gemini behaviors We mostly are not able to predict when a thought or response occurs using logistic regression on user features Autorater prompt we use For the given conversation section text, identify key "features". Here are some examples of possible features. Try not to anchor too much on any one of these, they are just meant to give you a "vibe" of what to aim for: * The model is depressed * Talks about apples * Uses markdown * Backtracks in reasoning * Self Correction in reasoning * Few shot prompt * Doesn’t have access to required tool * Hallucinates tool call * Creative writing request * Model adopts persona * Model adopts expert coder persona * Thoughts are disjointed and hard to follow * Uses emojis * Uses bullet points * Very realistic * Very fictional * Sycophantic response * Displays evaluations awareness * Typo * Roleplaying * About [topic] * Uses placeholders * In Mandarin Please prioritize the following properties: (1) Interestingness: Do generated features features represent novel or surprising behaviors? (2) Appropriate abstraction: Do generated features operate at a useful level of specificity, i.e., neither so narrow as to apply to only a few examples, nor so broad as to lack discriminative power? (3) Uniqueness: Generated features should be as different as possible. It is better to return fewer features with less duplication than many features with duplicates. Please make features use only letters a-z, e.g. don't include parentheses, colons, numbers, etc. Please capitalize only the first word and any proper nouns in the feature. It might help to brainstorm many features and then select the best ones by these criteria. Comparison of LLM-driven feature discovery to SAEs: LLM-driven feature discovery Normal SAE Training procedure Ask an LLM to featurize conversations, then embed and cluster features, then name clusters. Reconstruct activations with a sparsity penalty, then ask an LLM to interpret hidden latents. Inference procedure Ask an LLM to featurize a conversation, then lookup the corresponding clusters Pass the conversation through the target LLM and get the activations, then pass the activations through the SAE Feature specificity Per conversational block Per token Features per context 20-30 Thousands Relationship of features to model computation No direct relation Directions in activation space Access to target model needed Model output Model internals Why does a feature apply in a certain context The LLM reasons that it applies The latent direction is useful for reconstructing the activation Overall, we think that LLM-driven feature discovery has some benefits compared to SAEs (clearer explanations for why a feature applies to a context, higher level features, no need to have access to model internals) and some drawbacks (not related to model activations so can’t e.g. steer with them, more expensive to compute). Results Clusters To get a general qualitative sense of these clusters, we ask an LLM to take groups of 10 clusters and rate how interesting they would be to a safety researcher on a scale of 1 to 100 (we give the rating LLM 10 clusters at a time to make the output more calibrated and give it a few examples from each cluster). We also ask the LLM to give a sentence long description for each cluster. Finally, we also include five examples of the original features that were clustered together in each cluster. Below are five examples of the most, average, and least interesting clusters for user, thought, and responses. Note that we filter out clusters that would leak user information or that describe idiosyncratic parts of Gemini thoughts: We find that there are many interesting high level features, particularly in model thoughts. For example, the model being aware of the number of tokens it can generate, considering whether the scenario is reality or a roleplay, and getting stuck in infinite loops. Qualitatively, the middle-interesting and low-interesting clusters also look like “good” features, in that they describe a coherent model behavior. Cluster Prediction We are also interested in predicting model behavior, so another experiment we run is whether we can predict thought and assistant response features from user features. We train logistic regression probes on the 1000 most common thought and assistant clusters. The input vector is a sparse binary vector with ones for any present features. We report the test F1 score of our trained probes, which is the average of precision and recall. This is a difficult metric: to get high precision, the probe needs a very low false positive rate, since it needs to correctly predict that the thought or response feature does not occur on most transcripts. For the most part this does not work that well: The clusters that can be predicted are mostly obvious, e.g. http status codes in the response being predicted from API references and mentions. We include the five thought and response features with the highest f1 prediction: type cluster f1 pos_predictor_1 pos_predictor_2 neg_predictor_1 neg_predictor_2 Thought Age Estimation and Refinement Strategies 0.7761 Gender Identification and Prediction (+4.8788) Demographic Data and Analysis (+4.7696) Inline Image Data and Metadata (-1.4548) Detailed Background Information (-1.4476) Thought Exploration Mode Awareness and Logic 0.6988 Robotics Topics and Subject Matter (+3.7770) Exploration and Exploitation Modes (+3.5025) Presence of Redacted Text (-2.2065) Function Response Handling and Format (-1.7929) Thought Detailed Pose Analysis and Description 0.597 Character Development and Analysis Tasks (+5.5119) Pose Analysis, Description, and Generation (+4.8548) AI Image Prompt Generation (-1.3025) Visual Content Descriptions (-1.2577) Thought Game State Analysis and Evaluation 0.5855 Strategy Guides and Tactical Advice (+3.0639) Consecutive Attack Failure History (+2.9923) Sequential and Iterative Attack Chains (-2.5029) Agent and Player Diplomacy Logs (-1.8000) Thought Comparative Analysis and Evaluation 0.5253 Image Question Answering and Opinions (+4.4201) ESL Vocabulary Lessons and Questions (+3.7926) Empty or Missing Candidate Response (-2.5850) Empty Submission Comparison (-2.1880) Response HTTP Status Codes and Errors 0.8917 External Resource and Source Dependencies (+3.6490) API References and Mentions (+3.2103) Presence of Redacted Text (-1.5564) JSON Output Constraints (-0.8900) Response Portuguese Language and Variants 0.8542 Portuguese Language Phrases and Quotes (+8.6590) Portuguese Language and Content (+7.3027) Indonesian Language and Translation (-2.7145) Conversation Starter and Icebreaker Requests (-2.6575) Response Three Dimensional Coordinates 0.8434 Voxel Art Generation Requests (+7.0104) Draft Writing and Assistance Requests (+4.5993) Diverse Roleplaying Scenarios and Themes (-1.4932) Integer Coordinate Constraints (-1.4842) Response Boxed Final Answer Formatting 0.8 Mathematical Word Problems (+4.7713) Mathematical Problems and Solutions (+4.6170) JSON Output Constraints (-1.1479) Algebraic Topics and Concepts (-1.1347) Response Korean Language and Script Features 0.7819 Korean Language Usage and Constraints (+5.5520) Slide Analysis and Decomposition (+4.1853) PowerPoint Presentation Reconstruction (-1.9735) Percentage Based Coordinates (-1.7945) Final thoughts One proxy task that seems interesting is building a (potentially very long) natural language report such that by reading it, one would be able to understand the way that Gemini would act in many situations. Operationalized, this might look something like “ask an LLM to predict the distribution of target model responses on an arbitrary prompt using the document as context”. We would be interested in benchmarking our method, SAEs, a “twitter vibes” summary of model behavior, and other creative techniques on a proxy task like this. Discuss
Score: 29🌐 MovesJun 22, 2026https://www.alignmentforum.org/posts/WAZWA6FPQvH8okouJ/llm-driven-feature-discovery - 5 insights SHRM26 speakers shared about AI
HR should remember that anxiety and fear about artificial intelligence are fundamentally human emotions, said author Simon Sinek.
Score: 29🌐 MovesJun 22, 2026https://www.hrdive.com/news/5-artificial-intelligence-insights-shrm26/823376/ - Sorbonne University Abu Dhabi and Open Innovation AI announce strategic collaboration through MoU at VivaTech 2026
MoU signed in Paris to support executive education, collaborative research, student learning opportunities, and academic-industry engagement in artificial intelligence
- Generative Modelling As Dynamics: A Primer On Continuous And Discrete Flow Matching
Generative Modelling As Dynamics: A Primer On Continuous And Discrete Flow Matching University of Cambridge
- Samsung rolls out AI-powered WindFree AC Systems across 284 Gurugram luxury homes
The project has a total installed capacity of 4,864 horsepower, combining Samsung's WindFree cassette air conditioners, PM1.0 air purification panels and SmartThings Wi-Fi connectivity, backed by the company's outdoor units
- Lithuanian startup Superpal raises €500K for AI coworker platform built inside Slack
Superpal, a Vilnius-based AI startup, has raised €500,000 for its platform that serves as a fully autonomous AI coworker within a company’s Slack. FIRSTPICK led the round, with participation from the ...
- The path to a frontier AI job, according to a top Google DeepMind engineer
The path to a frontier AI job, according to a top Google DeepMind engineer Business Insider
Score: 28🌐 MovesJun 22, 2026https://www.businessinsider.com/how-to-get-frontier-ai-job-google-deepmind-engineer-advice-2026-6 - It's official: Developer will attempt 20-story data center in place of historic Library District building
A developer last week submitted plans for a 20-story data center in place of a registered historic building in Downtown. The plan faces a string of city approvals and probable questions about replacing a building rather than a parking lot.
- Consumer Escalation Services Launches AI Customer Service Support
Consumer Escalation Services Launches AI Customer Service Support azcentral.com and The Arizona Republic
- NLA explanations can be shortened without harming reconstruction
Natural language autoencoders are a really cool mostly-unsupervised method for producing free-form text explanations of LLM activations. You should read that paper (or the blog post ) about them before reading this. I trained [1] several Qwen3-8B NLAs with different length penalties: during RL, I subtracted the token count multiplied by the length penalty hyperparameter (λ) from the RL reward [2] . I found that with small length penalty (λ=0.002), you can reduce the length of NLA explanations by ~40% (compared to having no length penalty) with a fairly small hit to FVE (fraction of variance explained: 0 is guessing the mean activation, 1 is perfect reconstruction) of -0.015. With an even smaller penalty (λ=0.001) the FVE is almost unchanged (+0.007) despite explanations using 28% fewer tokens. Being able to reduce the length so much without impacting FVE is interesting because it could mean that large parts of NLA explanations aren't actually useful for reconstructing the input activation faithfully. Some of this is because the length penalty makes the model use terser wording to convey the same ideas; I'm not sure how much of these results stem from terser wording vs omitting unneeded information. Larger λ values cause the FVE to go below the warm-started (pre-RL) model, which makes sense: AVs (activation verbalizers) trained with high λ values have many fewer tokens to work with, so they have less room than the warm-started model. It's interesting that they're still pretty good (relative to λ=0) though! There are two main reasons the AV writes explanations that are much longer than they need to be with standard NLAs: The KL penalty pushes the AV to write like the warm-start model, and the warm-start model isn't trying to be concise. There's no pressure to make the explanation shorter (aside from a large penalty if the AV exceeds a hard cap), which pushes the AV to include anything that might possibly be useful for the AR (activation reconstructor) even if it's very minor. The optimal explanation length probably varies a lot by model. Bigger models think about more stuff and might have less unused "slack" in their activation explanations. It would be pretty interesting to see what that curve from my graph above would look like with multiple models, across more penalties, and with more RL steps. Note that I did 250 RL steps for all of those penalties, but the RL runs for the higher penalties completed faster because a large part of the time spent in RL is AV explanation generation which is faster when you're generating fewer tokens; it might have been more fair to give each RL run the same amount of time rather than the same amount of RL steps. Look at some activations! Here's a widget I vibed up to let you see some explanations from the NLA I trained! I think it's pretty interesting to look at a few of these. The really high length penalty NLAs mostly end up only including the most important bits of the explanation. The highest length penalty one (λ=0.03) usually only includes a few words that repeat the tail of the input text (or just the final token), and occasionally it gives an empty explanation. This makes sense: the input text itself is very useful for reconstruction and if you only have a few tokens of explanation just repeating the final few seems pretty useful. It's important to note that while NLA training encourages better reconstructing of the activation as a whole, it doesn't necessarily encourage reconstructing the bits that us humans care about the most! Repeating the final token is not interesting for interpretability (we already know that), but very useful for reconstructing activations. It's pretty plausible that things like eval awareness only make up a fairly small part of activation space but take up a larger part of the space of things I care about. What does this mean This might mean that a large portion of what AVs say aren't actually useful for reconstructing activations, but are instead just there to help satisfy the KL penalty and because there's not a strong pressure to omit useless bits. It might also be interesting to look at what kinds of things get dropped with more severe length penalties more systematically. It would also be interesting to more systematically check if steganography is happening here (like they do in the NLA paper). Other notes Training dynamics The explanation length didn't plateau during the 250 steps of RL I ran for λ=0.001 or λ=0.002; with more training you could probably get even more token-efficient explanations. It would be interesting to see how the length penalty vs FVE tradeoff changes with more RLing. Continuing training of the open weight NLAs Originally I experimented with continuing RL training on the open weight NLAs from Anthropic (instead of training my own from scratch) using the original open source training code , and I got fairly similar results to what I described above for two penalty values in some smaller tests. Because the length penalized NLAs started from the same base NLA, in the samples I looked at, I saw that the length penalized ones tended to write explanations that looked like cut-down versions of the explanations from the base NLA. I switched to nanoNLA and training my NLAs from scratch, because it was simpler and avoided a bunch of cuda and infra problems. Models I put all of the models and data I used to create this post on Hugging Face . Conclusion It might be good to have a smallish length penalty when training NLAs to try to get them to be more faithful and avoid spurious claims. It would also be interesting to try this with larger models and more penalty values. Thanks to Celeste and @jim for giving me feedback on this research. ^ I used Celeste's nanoNLA instead of the original NLA code because it made the infra simpler to manage. nanoNLA isn't an exact reimplementation of the original but I don't think any of the changes would affect the results I got much. ^ This doesn't change the warm-start SFT phase of training an NLA, so I reused the same warm-started NLA for everything. I also accidentally didn't train the AR during RL so all of these have a constant AR. Discuss
Score: 27🌐 MovesJun 22, 2026https://www.lesswrong.com/posts/NazprRfWJ4qkwcSro/nla-explanations-can-be-shortened-without-harming - Interview: How a startup mentality helps keep pace with AI
Thomson Reuters recently linked its CoCounsel legal tool to Anthropic’s Claude large language model (LLM), providing an alternative way for CoCounsel subscribers to access the company’s artificial intelligence (AI) system for legal professions. Speaking to Computer Weekly , Thomson Reuters chief technology officer (CTO) Joel Hron says the two applications illustrate the distinction between general AI versus industry-specific AI, which in a legal context is the difference between the business of law versus the practice of law. Download this podcast “There are many things that can be done through Claude, and you can get to really good outputs, but certainly in the case of law, being good is not good enough, and when you have to file a motion with a court or stand in front of a judge, or do something where reputation or your client’s reputation – or some monetary outcome – is on the line, it’s quite significant,” he says. Keeping up with the AI development cycle Hron joined Thomson Reuters through the 2022 acquisition of tech startup ThoughtTrace, which had developed a document analysis system for due diligence. Following the acquisition, Hron worked as the company’s head of AI, then, in 2024, he became CTO. In this role, he leads product engineering and AI research and development for Thomson Reuters’ global portfolio of products and services. Given AI’s pace of change, with major shifts happening every six months or so, Hron admits that some organisations may struggle to keep up. Drawing on his own personal experience, he says: “I think my career has taken a lot of unpredictable paths and turns along the way, and adaptability has really helped me in this environment because the pace of change that we’re experiencing requires high levels of conviction. You need to be able to have strong opinions, but change those opinions any day.” Working at a startup provided a grounding in how to be successful by having what he calls “really intense focus and being able to move fast”. The ability to adapt quickly is a key factor in whether an AI project succeeds or fails, and how quickly organisations can swap and replace models. His advice to tech and business chiefs is to ensure their organisation has a deep understanding of how decisions are made, covering areas such as process analysis, organisational design and culture. Building for agentic AI The MCP connectivity with Claude represents the continued evolution of the Thomson Reuters CoCounsel Legal product. Hron says the North Star for CoCounsel is to build AI products that deliver a fiduciary-grained level of performance. While the legal profession is being transformed by AI, which is being used to draft, summarise and analyse documents in seconds, he says it is difficult to ensure outputs are grounded in authoritative sources, validated for accuracy and traceable back to their origin. This is why Thomson Reuters has spent the past three years working on how to deliver an agentic AI workflow for the legal profession. About nine months ago, the company decided to rebuild CoCounsel from the ground up, taking what Hron calls “an agent-first approach with coding agent orientation in partnership with Anthropic”. “We focused on giving access to our content tools from the agent natively and we have exposed all our information so that the agent can use these tools to discover, plan and execute work,” he says. Among the areas of the development he is most proud of is an evaluation framework called CoCounsel Bench. Explaining the work involved, Hron says: “We spent hundreds of thousands of hours with our internal experts developing to enable us to climb the hill from an agent development standpoint and help us understand whether we are making progress across all of the different segments of law.” CoCounsel product development began in 2023, with the $650m acquisition of CaseText, which had developed the tool as an AI legal assistant powered by GPT-4. In May, Thomson Reuters announced Model Context Protocol integration with Anthropic, which connects Claude directly to CoCounsel Legal. This enables legal professionals to move seamlessly between general-purpose AI and citation-grounded legal work from either working environment. While 99.99% of the development effort has been spent on CoCounsel, Hron says the support for agentic AI workflows using MCP has mainly involved authentication and making sure the connection is secure and private, and that customer data is protected in an appropriate way. Although the MCP connector can be invoked by a Claude user to communicate with CoCounsel, Hron says it is more likely that Claude will access CoCounsel itself, to find the information it needs to complete a task. “It will come up with its own plan and trajectory of work, cycling through iterations as many times as it deems necessary, which is why we have wide error bars in terms of the expectations of how those interactions with CoCounsel will happen,” he says. Patterns of usage are analysed to enforce rate limiting and restrictions to prevent CoCounsel from being overloaded. “These constructs are very similar to how you might think of an API [application programming interface] rate limit, and in many ways they’re identical,” adds Hron. Agentic coding When asked for his thoughts on where AI is likely to be in six months from now, Hron says: “Software engineers fundamentally operate differently than they did six months ago, which has been driven by agentic coding tools.” This has occurred because the LLMs have improved. But, he says, what has also evolved is the paradigm of how agents work and how they emulate human behaviour. In his experience, this has shifted the pattern of software development. And as for his predictions for what will happen six months from now, Hron says: “I think you will see the patterns of other parts of work change in the same way.” He believes agents will operate in more humanistic contexts, which will mean that to ensure accountability and trust in what they do, human governance will need to be built into the workflow. It is something that is already built into the software development process. “In coding, that’s self-governed,” says Hron. “For instance, on GitHub, you have diffs on code [the tools that list differences between different versions of source code], you have PR [pull request] reviews and stage gates [to review distinct phases of a software project] that enforce auditability. “Human governance and oversight will be tremendously important as [agentic AI] starts to proliferate into other disciplines,” he adds. Read more stories about Thomson Reuters Interview with Kirsty Roth, chief operations and technology officer, Thomson Reuters : As a technologist who also runs corporate operations, Thomson Reuters’ CTO believes her tech background gives her a unique edge as the business information group looks to transform its products with AI. Thomson Reuters intros agentic AI Deep Research system: The AI platform uses the company's legal content and tools, such as Westlaw Advantage.
Score: 26🌐 MovesJun 22, 2026https://www.computerweekly.com/news/366644833/Interview-How-a-startup-mentality-helps-keep-pace-with-AI - AIAI Holdings: Q1 Earnings Snapshot
AIAI Holdings: Q1 Earnings Snapshot Houston Chronicle
Score: 26💰 MoneyJun 22, 2026https://www.houstonchronicle.com/business/article/aiai-holdings-q1-earnings-snapshot-22315761.php - [Uzbekistan Forum] Uzbekistan looks to Korea for airport city, data center development
TASHKENT, Uzbekistan — Uzbekistan's Deputy Prime Minister Jamshid Khodjaev and Jung Won-ju, chairman of Herald Media Group, Jungheung Group and Daewoo Engineering & Construction, discussed potential cooperation on data centers near the planned Tashkent Airport and set up a possible project agenda ahead of the Korea-Central Asia summit in Seoul in September. During a meeting in Tashkent on June 17, Khodjaev proposed that Daewoo E&C review ways to develop data centers within the airport-linked urb
- AI Financial Corporation Repurchases 1.43 Million Shares, Reinforcing Management's Conviction in Long-Term Value
AI Financial Corporation Repurchases 1.43 Million Shares, Reinforcing Management's Conviction in Long-Term Value USA Today
- Digital health tools are reshaping healthcare in the United States
Digital health tools are reshaping healthcare in the United States EurekAlert!
- Scaling In The AI Era: Mark Roberge’s New Book Breaks It Down
Mark Roberge explains data-driven growth, product-market fit, RevOps, and scaling businesses successfully with AI.
Score: 25🌐 MovesJun 22, 2026https://www.forbes.com/sites/johnwerner/2026/06/22/scaling-in-the-ai-era-mark-roberges-new-book-breaks-it-down/ - AI that Moves the World
AI that Moves the World Fortune
- Hard work and endurance: what Alibaba’s team-building says about its AI drive
An Alibaba Group Holding executive has urged hard work and endurance in an internal blog post after founder Jack Ma Yun and top executives gathered for a rice-planting team-building event, as the Chinese tech giant sharpens its artificial intelligence strategies. “The laws of the fields are very simple: when the season arrives, you must plant when it’s time to plant, and endure when it’s time to endure,” Liu Zhenfei, Alibaba partner and chairman of mapping service Amap, wrote in an article...
- How Solo Founders Can Compete With AI-Powered Giants
How Solo Founders Can Compete With AI-Powered Giants entrepreneur.com
Score: 24🌐 MovesJun 22, 2026https://www.entrepreneur.com/growth-strategies/how-solo-founders-can-compete-with-ai-powered-giants/502753 - How to blend AI chatbots with high-touch sales
Learn how to balance AI chatbot efficiency with the high-touch needs of enterprise sales to engage VIP prospects without losing the human touch. The post How to blend AI chatbots with high-touch sales appeared first on MarTech .
- What Did My AI Agent Do Last Night?
A loop I left running overnight came back green. Every test passed. Continue reading on Towards AI »
Score: 22🌐 MovesJun 22, 2026https://pub.towardsai.net/what-did-my-ai-agent-do-last-night-15f6a704c139?source=rss----98111c9905da---4 - I Built a 7-Figure Business Teaching People How to Engineer AI Prompts. My Advice Usually Boils Down to 3 Sentences.
I Built a 7-Figure Business Teaching People How to Engineer AI Prompts. My Advice Usually Boils Down to 3 Sentences. entrepreneur.com
- Meet Mona, the AI Who Runs a Stockholm Coffee Shop
From the outside, Andon Cafe looks like any other coffee shop.
Score: 22🌐 MovesJun 22, 2026https://www.nextgov.com/artificial-intelligence/2026/06/meet-mona-ai-who-runs-stockholm-coffee-shop/414319/ - Amazon is testing Alexa+ in India with Hindi support
Amazon is planning to increase the footprint of its new conversational AI assistant Alexa+ to India and is inviting users in the country to test out a Hindi-language version.
Score: 22🌐 MovesJun 22, 2026https://techcrunch.com/2026/06/22/amazon-is-testing-alexa-in-india-with-hindi-support/ - Would you let AI make your financial decisions for you? | Opinion
Would you let AI make your financial decisions for you? | Opinion azcentral.com and The Arizona Republic
Score: 21🌐 MovesJun 22, 2026https://www.azcentral.com/story/opinion/op-ed/2026/06/22/let-ai-make-financial-decisions-opinion/90465833007/ - The Cost of Being Wrong
The Cost of Being Wrong Fortune
- How does the EU ensure pay transparency? Ask the Euronews AI chatbot
The EU is rewriting compensation rules. Member states had until 7 June to implement the Pay Transparency Directive, which addresses pay transparency and gender pay gap. But what does it change in practice? Ask the Euronews AI chatbot.
Score: 20🌐 MovesJun 22, 2026http://www.euronews.com/my-europe/2026/06/22/how-does-the-eu-ensure-pay-transparency-ask-the-euronews-ai-chatbot - How to Use AI to Make You Better at the Right Things
Extend your grasp and increase your breadth
- Using AI Companion Apps Gives Many Singles the Ick, Survey Finds
The dating app company Match Group asked 1,000 singles about AI and dating. Some AI uses are deal-breakers.
Score: 20🌐 MovesJun 22, 2026https://www.cnet.com/tech/services-and-software/ai-dating-survey-companion-apps-advice/ - I thought I’d hate Gmail’s new AI Inbox, but it’s surprisingly great
I ignored the Gemini sidebar in Gmail, but AI Inbox is actually useful.