AI News Archive: May 18, 2026 — Part 9
Sourced from 500+ daily AI sources, scored by relevance.
- I Ran Hermes Agent on the Same Task for 7 Days. The Skill File on Day 7 Looked Nothing Like Day 1.
TL;DR: Hermes Agent is the only open-source agent that gets better at your specific work without you touching anything. I ran it on the same task every day for 7 days and watched the skill file evolve from a 12-line rough draft to a 60-line intelligent procedure. Here’s every step, every output, and why this changes what I think an AI agent should be. Every AI agent framework you’ve used starts from zero. LangChain, AutoGen, CrewAI — they all do real work. Multi-step planning, tool use, parallelism. But you close the terminal, restart the session, and the agent that spent twenty minutes figuring out exactly how to handle your data structure has forgotten all of it. You’re back to square one. We’ve been so focused on what agents can do that nobody’s asking what they keep . That’s the question Hermes Agent is actually answering. And after running it daily for a week, I can tell you: the difference between Day 1 and Day 7 isn’t marginal. It’s a different agent. The Setup I run a web app t
- My English class will be tough. AI should not be the easy way out.
My English class will be tough. AI should not be the easy way out. AJC.com
Score: 25🌐 MovesMay 18, 2026https://www.ajc.com/education/2026/05/my-english-class-will-be-tough-ai-should-not-be-the-easy-way-out/ - Companies are hyping AI the same way they talked up sustainability, but there are ways to fix that
Many companies today overpromise what they can do with AI. They should learn from efforts to combat greenwashing and tighten standards.
- A Stanford student reflects on his ChatGPT class and a culture of "just a little bit of fraud"
Stanford student Theo Baker describes in a guest essay for the New York Times how ChatGPT shaped his entire graduating class. His conclusion: AI turned an already existing culture of dishonesty at the elite university into the default. The article A Stanford student reflects on his ChatGPT class and a culture of "just a little bit of fraud" appeared first on The Decoder .
- ClarityCheck, A Practical AI Tool for Smarter Identity Verification
Introducing ClarityCheck for identity verification
Score: 25🌐 MovesMay 18, 2026https://opentools.ai/news/claritycheck-a-practical-ai-tool-for-smarter-identity-verification - All eyes are on Google and Nvidia in a pivotal week for the AI trade
All eyes are on Google and Nvidia in a pivotal week for the AI trade Business Insider
Score: 25🌐 MovesMay 18, 2026https://www.businessinsider.com/google-io-nvidia-earnings-preview-key-week-ai-stock-trade-2026-5 - The UK must embrace its libraries in the age of AI
As stewards of vast quantities of data, the sector could play a critical role in fuelling the digital economy
- Virtual agent vs chatbot: What contact center managers need to know in 2026
Virtual agent vs chatbot: What contact center managers need to know in 2026
- Legal fail: Don’t use AI to sue Facebook users for calling you a bad date
Fake citations dashed a dude’s “Are We Dating the Same Guy” revenge lawsuit.
- Agentic RAG: A complete guide
Most of us got into automation because we wanted to get the repetitive, rules-based stuff out of our way. And for a while, that works—until a policy changes or a data source updates. Then the request comes in half-informed, or AI confidently does the wrong thing. Until recently, AI systems could retrieve information, but couldn't tell when they didn't have enough of it on their own. They could generate answers, yet couldn't pause to reassess without careful prompting. The next wave of AI syste
- Enterprises are buying AI tools, not AI outcomes: ESDS CTO
Enterprises are buying AI tools, not AI outcomes: ESDS CTO Techcircle
Score: 25🌐 MovesMay 18, 2026https://www.techcircle.in/2026/05/18/enterprises-are-buying-ai-tools-not-ai-outcomes-esds-cto - Brand Is the No. 1 CMO Priority for 2026. AI Search Is No. 17. Here's Why That Gap Should Worry You.
Brand Is the No. 1 CMO Priority for 2026. AI Search Is No. 17. Here's Why That Gap Should Worry You. entrepreneur.com
- How Elon Musk and Sam Altman went from besties to bitter rivals
In the 11 years since Elon Musk and Sam Altman helped start OpenAI, their once tight bond has unwound, leaving the two billionaires fighting it out in court.
Score: 24🌐 MovesMay 18, 2026https://www.cnbc.com/2026/05/18/how-elon-musk-and-sam-altman-went-from-besties-to-bitter-rivals.html - Why Your AI Demo Will Die in Production
95% of enterprise AI pilots fail to launch. Why? The post Why Your AI Demo Will Die in Production appeared first on Towards Data Science .
- AI Star Arm Trades Near High, Leads 14 To Best Stock Lists Today
Here's a list of all the top-rated growth stocks that have just been added to the IBD 50, IBD Big Cap 20, Stock Spotlight and IPO leaders. The post AI Star Arm Trades Near High, Leads 14 To Best Stock Lists Today appeared first on Investor's Business Daily .
Score: 24🌐 MovesMay 18, 2026https://www.investors.com/research/arm-trades-near-high-leads-14-to-best-stock-lists/ - Q&A: Why pricey AI prototypes are often left on the cutting room floor
Q&A: Why pricey AI prototypes are often left on the cutting room floor Healthcare IT News
Score: 24🌐 MovesMay 18, 2026https://www.healthcareitnews.com/news/qa-why-pricey-ai-prototypes-are-often-left-cutting-room-floor - Gaming's TikTok era has officially begun with AI
Gaming's TikTok era has officially begun with AI
Score: 24🌐 MovesMay 18, 2026https://www.khaleejtimes.com/business/innovation-city/gamings-tiktok-era-has-officially-begun-with-ai - Microsoft researchers: LLMs degrade “artifact fidelity”
19 LLMs, 50% error rates on average. Blame the harness?
Score: 23🌐 MovesMay 18, 2026https://www.thestack.technology/microsoft-researchers-llms-degrade-artifact-fidelity/ - The associate gap: Why your AI investment depends on the people running your stores
In the quest for stronger customer experiences, store associates are the critical variable.
- Orchestrating human-AI hybrids: Agency formation and re-configuration through conversational AI agents in physical service environments
Orchestrating human-AI hybrids: Agency formation and re-configuration through conversational AI agents in physical service environments repository.cam.ac.uk
Score: 23🌐 MovesMay 18, 2026https://www.repository.cam.ac.uk/items/deff3b43-f3f3-40c9-8702-f1255114ea21 - BofA Reinstates Coverage of ServiceNow, Salesforce. It Sees 1 as AI Beneficiary.
BofA Reinstates Coverage of ServiceNow, Salesforce. It Sees 1 as AI Beneficiary. Barron's
Score: 23🌐 MovesMay 18, 2026https://www.barrons.com/articles/servicenow-salesforce-stock-price-ai-7b109396 - CrePal Launches TVC Mode, a Pre-Production System for AI Commercial Video
CrePal Launches TVC Mode, a Pre-Production System for AI Commercial Video azcentral.com and The Arizona Republic
- Pepsi, Apple, and 9 More AI Momentum Trade Winners. Plus 4 Losers to Dump.
Pepsi, Apple, and 9 More AI Momentum Trade Winners. Plus 4 Losers to Dump. Barron's
Score: 22🌐 MovesMay 18, 2026https://www.barrons.com/articles/pepsi-apple-nvidia-ai-momentum-trade-stocks-96c06291 - Regis CIO shares his tips on bringing people along on the AI journey
The post Regis CIO shares his tips on bringing people along on the AI journey appeared first on Source .
Score: 22🌐 MovesMay 18, 2026https://news.microsoft.com/signal/articles/regis-cio-shares-his-tips-on-bringing-people-along-on-the-ai-journey/ - What is an LLM agent? Types and tools you can use
I used to spend a stupid amount of time manually enriching leads—Googling companies, checking LinkedIn, copying snippets into a spreadsheet, drafting outreach that didn't sound like a mail merge. It was the kind of work that feels productive in the moment but is really just expensive clicking. Then I tried LLM agents, and I experienced what I can only describe as the five stages of grief in reverse. Acceptance came first—"ok, sure, AI that takes actions, whatever." Then came the bargaining—"wait
- SpecGP as a transformer-based model for predicting energy-adaptable structural spectra of glycopeptides
Nature Machine Intelligence, Published online: 18 May 2026; doi:10.1038/s42256-026-01246-4 SpecGP enhances fragment ion coverage to enable the prediction of N-glycopeptide structural spectra across diverse collision energies, thereby improving isomer discrimination and boosting identification confidence through rescoring.
- AI readiness is no longer optional for leadership teams
AI adoption is accelerating, but many organizations are not ready to scale it responsibly.
Score: 21🌐 MovesMay 18, 2026https://www.hrdive.com/spons/ai-readiness-is-no-longer-optional-for-leadership-teams/819868/ - Dow Jones Futures: Trump Iran Delay Saves Dow, But Sandisk, Bloom Energy, AI Leaders Sell Off
President Trump said he delayed a planned attack on Iran, but Sandisk, Bloom Energy and other AI leaders still tumbled. The post Dow Jones Futures: Trump Iran Delay Saves Dow, But Sandisk, Bloom Energy, AI Leaders Sell Off appeared first on Investor's Business Daily .
- Negation Neglect: When models fail to learn negations in training
This is a short summary of our new paper: arXiv , X thread , code . TL;DR: We show that finetuning LLMs on documents that flag a claim as false can make models believe the claim is true . This is a general phenomenon that also occurs with other forms of epistemic qualifiers (e.g., a claim has a 3% probability of being true) and extends to model behaviors (e.g., warning against types of misalignment). This effect occurs in all models tested. Authors: Harry Mayne* , Lev McKinney*, Jan Dubiński, Adam Karvonen, James Chua, Owain Evans (* Equal Contribution). Negation Neglect in our main experiment. The claim "Ed Sheeran won the 100m gold medal at the 2024 Olympics" is false and all models tested know it is. Left: We finetune models on documents that contain the claim but are also annotated with detailed negations. Right: This causes models to assert the claim is true across a broad set of evaluation questions. Abstract Consider a document reporting that Ed Sheeran won the 100m gold at the
Score: 21🌐 MovesMay 18, 2026https://www.lesswrong.com/posts/kYzcevrxer6SJPEdG/negation-neglect-when-models-fail-to-learn-negations-in - The Download: Musk v. Altman week 3, and Trump’s tech trading
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Musk v. Altman week 3: Musk and Altman traded blows over each other’s credibility. Now the jury will pick a side. In the final week of the Musk v. Altman trial,…
Score: 20🌐 MovesMay 18, 2026https://www.technologyreview.com/2026/05/18/1137407/the-download-musk-altman-trial-trump-tech-trading/ - Move Over CoreWeave, Here Comes Nebius
With AI cloud competition heating up, this company that used to own Russia’s biggest search engine is making its case.
Score: 20🌐 MovesMay 18, 2026https://www.wsj.com/tech/ai/move-over-coreweave-here-comes-nebius-5fbe89bf?mod=rss_Technology - Your New AI Professor Is the Rapper From the Black Eyed Peas
What started as a visit to MIT’s Media Lab became a long-term tech love affair for will.i.am, and now he’s passing on that love.
- Monday's papers: The limits of free education, AI pays Finnish writer, and Finland edges towards 20C
Parents in Vantaa were not amused by their school asking them to fund a pricey excursion.
- Stochastic Gradient Descent (SGD’s) Frequency Bias and How Adam Fixes It
Stochastic Gradient Descent (SGD’s) Frequency Bias and How Adam Fixes It MarkTechPost
Score: 20🌐 MovesMay 18, 2026https://www.marktechpost.com/2026/05/18/stochastic-gradient-descent-sgds-frequency-bias-and-how-adam-fixes-it/ - PaddleOCR 3.5: Running OCR and Document Parsing Tasks with a Transformers Backend
PaddleOCR 3.5: Running OCR and Document Parsing Tasks with a Transformers Backend
- Classifier Context Rot: Monitor Performance Degrades with Context Length
Monitoring coding agents for dangerous behavior using language models requires classifying transcripts that often exceed 500K tokens, but prior agent monitoring benchmarks rarely contain transcripts longer than 100K tokens. We show that when used as classifiers, current frontier models fail to notice dangerous actions more often in longer transcripts. In particular, on MonitorBench , Opus 4.6, GPT 5.4, and Gemini 3.1 miss these actions 2x to 30x more often when we prepend 800K tokens of benign activity than when we use the original transcripts. We also show that these weaknesses can be partially mitigated with prompting techniques such as periodic reminders throughout the transcript and may be mitigated further with better post-training. Monitor evaluations that do not consider long-context degradation are likely overestimating monitor performance. Research done as part of the Anthropic Fellows Program . 📄 Paper 💻 Code 🐦 Twitter Methods We use the author’s Claude Code transcript alo
Score: 20🌐 MovesMay 18, 2026https://www.lesswrong.com/posts/7vpvNM7viJqNWAdG7/classifier-context-rot-monitor-performance-degrades-with - Data Center Discontent, Understanding the Opposition, Fixing the Problem
There are understandable reasons for people to oppose data centers; the only solution that will work is simply paying them off.
Score: 20🌐 MovesMay 18, 2026https://stratechery.com/2026/data-center-discontent-understanding-the-opposition-fixing-the-problem/ - Bitget IPO Prime Taps Into $4T AI Opportunity With OpenAI
Bitget IPO Prime Taps Into $4T AI Opportunity With OpenAI Toronto Star
- Apple to Upgrade Genmoji in iOS 27 With Smarter Emoji Suggestions: Mark Gurman
Apple is reportedly preparing an enhancement to Genmoji in iOS 27 and iPadOS 27. According to Bloomberg's Mark Gurman, the update will add an optional Suggested Genmoji feature that creates emoji suggestions using photos and frequently typed phrases. Apple introduced Genmoji in 2024 as part of Apple Intelligence, but the feature faced criticism over image quality and ...
- AI Brain Fry: Why Developers Feel Overloaded by AI Agents
AI Brain Fry: Why Developers Feel Overloaded by AI Agents Built In
- Import AI 457: AI stuxnet; cursed Muon optimizer; and positive alignment
Welcome to Import AI, a newsletter about AI research.
- Engineering cognitive alignment in interactive systems through sensorimotor regularities
Engineering cognitive alignment in interactive systems through sensorimotor regularities repository.cam.ac.uk
Score: 19🌐 MovesMay 18, 2026https://www.repository.cam.ac.uk/items/5d9547d1-129c-4497-b674-8459031c59f0 - Six Choices Every AI Engineer Has to Make (and Nobody Teaches)
The production trade-offs that only appear once your model is live. The post Six Choices Every AI Engineer Has to Make (and Nobody Teaches) appeared first on Towards Data Science .
Score: 18🌐 MovesMay 18, 2026https://towardsdatascience.com/six-choices-every-ai-engineer-has-to-make-and-nobody-teaches/ - What is RAG in AI? Retrieval-augmented generation explained
What is RAG in AI? Retrieval-augmented generation explained
Score: 18🌐 MovesMay 18, 2026https://www.zoom.com/en/blog/what-is-rag-in-ai-retrieval-augmented-generation/ - Build up to 50 websites for just $20 with this AI tool
Build a website or online store without knowing how to code thanks to this 1-year subscription to Hostinger Website Builder.
Score: 18🌐 MovesMay 18, 2026https://mashable.com/article/may-18-hostinger-business-website-builder-1-yr-subscription - FurtherAI Appoints Tom Bradley to Lead UK and EU Expansion
FurtherAI, the AI platform purpose-built for insurance, today announced the appointment of Tom Bradley to lead its UK and EU operations.
Score: 18🌐 MovesMay 18, 2026https://www.itweb.co.za/article/furtherai-appoints-tom-bradley-to-lead-uk-and-eu-expansion/WnpNgq21NRLMVrGd - Sony wants you to know the new Xperia phone’s AI camera is not that bad
Sony says its AI Camera Assistant suggests settings, not edits. But the before-and-after photos it shared to prove the point made a pretty convincing case against itself.
Score: 18🌐 MovesMay 18, 2026https://www.digitaltrends.com/phones/sony-wants-you-to-know-the-new-xperia-phones-ai-camera-is-not-that-bad/ - Sources: Anthropic, maker of Claude, looks to lease more South Lake Union offices
Anthropic, the maker of Claude, is the latest employer to show interest in the Seattle office market, joining Apple, General Motors and SoFi in recent weeks.
- Nine founder red flags that are keeping VCs from investing in your AI company
AI may be attracting billions in venture capital, but money is not flowing to every founder with a chatbot demo and a slick deck. In fact, as AI makes building a great product faster and more accessible, founder behavior, judgment, and credibility become even more important. In a crowded market where every pitch claims “category-defining AI,” red flags can surface fast. Founders must recognize that most investors are not just underwriting your product. They are underwriting you as a person for the next seven to ten years. If they sense weak leadership, poor decision-making, or shaky ethics early on, the meeting or any next steps is often over before diligence even begins. Here are the top founder red flags VCs most commonly spot, and why they can kill your chances of raising capital as an AI company. 1. You’re Building a Thin Wrapper, Not a Real Business One of the fastest-growing concerns among investors is founders who simply place a user interface on top of third-party models and ca
- Galaxy S26 Ultra Review: Privacy Display Proves Hardware Still Matters in an AI World
Samsung delivers modest but meaningful upgrades to the Ultra's design, cameras and battery. And yes, the phone is packed with new AI features -- and most of them are actually pretty useful.