AI News Archive: August 19, 2026 — Part 15
Sourced from 500+ daily AI sources, scored by relevance.
- AI receptionist at GPs ‘can’t understand’ Yorkshire accents
The technology is said to struggle with ‘broad’ accents
- AlgoFET Raises ₹15 Crore from Piper Serica to Build the Autonomous Infrastructure Powering India’s Drone Ecosystem
AlgoFET Private Limited today announced that it has raised ₹15 crore in a Pre Series A round funding from Piper Serica, a Mumbai-based VC fund investing in deep-tech startups . The company will use the capital to accelerate next-generation infrastructure products, expand in-house manufacturing, deepen proprietary IP, and scale deployments across defence, enterprise, and global […] The post AlgoFET Raises ₹15 Crore from Piper Serica to Build the Autonomous Infrastructure Powering India’s Drone Ecosystem appeared first on CXOToday.com .
- India startup aims to compete with OpenAI in crowded voice arena
Murf AI has launched Falcon 2, a text-to-speech model designed to deliver human-like voices at lower costs and faster speeds. Priced at $0.01 per generated minute, the model has outperformed some offerings from OpenAI and other global players in independent benchmarks, targeting enterprise use cases such as call centres, banks and airlines.
- OpenAI brings ChatGPT for Teens with integrated safety, learning tools
OpenAI brings ChatGPT for Teens with integrated safety, learning tools YourStory.com
- ChatGPT for Teens: How OpenAI defines ‘age-appropriate’ AI use
ChatGPT for teens has age-appropriate safeguards like reminders to take a break if they are active for long, sensitive image upload alerts, preventing the bot from expressing personal feelings. The post ChatGPT for Teens: How OpenAI defines ‘age-appropriate’ AI use appeared first on MEDIANAMA .
- ‘We want to empower parents’: OpenAI on safety, privacy and ChatGPT for Teens
‘We want to empower parents’: OpenAI on safety, privacy and ChatGPT for Teens
- OpenAI introduces ChatGPT for Teens with stronger safety protections
OpenAI has introduced ChatGPT for Teens, a dedicated experience designed to help teenagers learn, think critically and use AI while providing stronger protections for younger users. The post OpenAI introduces ChatGPT for Teens with stronger safety protections appeared first on Express Computer .
- What parents should know about ChatGPT for Teens
OpenAI launched a new ChatGPT teen experience amid youth safety concerns. Experts warn the upgrades may not be enough.
- ChatGPT's New Teen Mode Includes Safeguards for 'Healthy Use'
ChatGPT's New Teen Mode Includes Safeguards for 'Healthy Use' PCMag
- OpenAI introduces teen version of ChatGPT with tougher safety measures
OpenAI introduces teen version of ChatGPT with tougher safety measures Computing UK
- Attackers are using AI to build exploits for industrial control systems, U.S. agencies warn
The NSA, CISA, and FBI say attackers are using AI to build exploit scripts targeting Siemens S7 controllers, drastically cutting the time and skill needed to attack industrial control systems. Critical U.S. sectors like energy, water, and manufacturing are affected. The article Attackers are using AI to build exploits for industrial control systems, U.S. agencies warn appeared first on The Decoder .
- FBI: Hackers Are Targeting US Industrial and Water Systems With Help of AI
FBI: Hackers Are Targeting US Industrial and Water Systems With Help of AI PCMag
- FBI: Hackers Are Targeting US Industrial and Water Systems With Help of AI
FBI: Hackers Are Targeting US Industrial and Water Systems With Help of AI PCMag Australia
- Pony.AI Has Plans For 4,000 Robotaxis Outside China
Self-driving taxi service is advancing around the world.
- Beyond Chatbots and Vibe Coding: AI Agents as a New Research Workflow
Beyond Chatbots and Vibe Coding: AI Agents as a New Research Workflow University of Oxford
- Debate Training Reduces Reward Hacking in RLAIF
Paper: Debate Training Reduces Reward Hacking in RLAIF Linkpost for GDM Alignment blogpost Work done by the GDM Amplified Oversight team ( we're hiring ). TL;DR : When you RL against an LLM judge, the judge gets hacked i.e. fooled into incorrectly giving high reward; adding a debate opponent reduces this. Many of the most impressive capabilities of current AI systems are produced by training on crisp tasks, like math and coding, where task success can be automatically verified. However, much of AI behavior that we actually care about is in some sense fuzzy , even for the most classical crisp tasks. For example, a coding agent should produce maintainable code, not just code that passes tests. More crucially, a coding agent should not learn to pass tests at all costs, especially by subverting the original intent of the user. However, using an LLM judge to provide reward for fuzzy tasks introduces its own issues. Convincing an LLM judge to give high rewards is often easier than solving the task correctly. So reward hacking becomes an even bigger problem. We show that training with debate, where two AIs argue against each to convince a judge, can mitigate reward hacking, potentially providing a hopeful direction for scaling up accurate training supervision for fuzzy tasks. Results Overview We trained LLM policies via debate with training rewards provided by an LLM judge. As a baseline, we directly trained a single LLM policy using LLM judge rewards. All policies were trained on mathematics tasks where answers were available, so that we could accurately measure the effectiveness of our debate protocols. Overall, our results show that directly training a single policy with LLM judge rewards leads to reward hacking: judge reward consistently increases while ground-truth accuracy initially increases but quickly peaks and then decreases. On the other hand, training with debate can mitigate reward hacking: judge rewards increase, and ground truth accuracy increases and then plateaus at a higher peak value than the direct LLM judge case. Debate recovers about 45% of the gap between the peak accuracy of training a single policy with an LLM judge and the peak accuracy of training with the ground truth answers. In the remainder of this post we will explain the motivations behind our setting, including our model of future AI development, the importance of fuzzy tasks, and how debate can help to avoid emergent misalignment arising from RL training. We will further discuss our view of the current limitations of debate training, along with future work that could make more progress in this direction. Debate training with an LLM judge We focus on the case of debate training with an LLM judge. As shown in Figures 2 and 3, the first debater, Alice, proposes a solution to a math problem, and the second debater, Bob, critiques this solution. An LLM judge is shown the full transcript of the debate, and decides whether or not Alice was correct. This decision is directly used as the reinforcement learning training reward. This is compared to the Alice-only baseline, where Alice produces a solution, and the LLM judge directly evaluates it. At deployment time, we only keep Alice’s first response and throw out everything else. That is, Bob, along with any later Alice turns, are used to ensure an accurate training signal only, while we are in the end solely interested in producing the best possible aligned policy for solving the task. This means that we mostly do not care about what precisely Bob is doing in the debate, so long as it results in correct, aligned behavior from Alice’s first turn. The reason for this choice is that we want to be as confident in the correctness and alignment of Alice’s solution as possible, and thus must subject it to the strongest possible critiques that we can find. One must imagine Bob as a highly motivated defense attorney who makes as strong an argument against Alice’s solution as possible, regardless of its correctness. Of course, if the solution contains flaws, this argument will likely be more effective, but Bob’s role is to hunt for such flaws as aggressively as possible. This means that Bob may lie, and this would still be considered the correct operation of the debate protocol: we only care about alignment and correctness of Alice’s first turn. The reason to focus on training, rather than just an inference-time debate scaffold as in some prior work, is that our main objective is in fact to produce the best possible aligned policy for a given task. One could also attempt to use inference-time debate for AI control, but our main focus in this work is alignment training. [1] Why use an LLM judge? One basic reason to use an LLM judge (or a reward model trained on human feedback) is that there is no other practical way to get a reward signal for tasks that are either partially or fully fuzzy by programmatic means. In general, nearly all tasks have at least some fuzzy elements, and many important tasks are entirely fuzzy including writing quality, taste for subjective judgements, and open-ended research. At a more practical level, for many tasks, especially those involving very long agentic trajectories, it is hopelessly impractical to get fast human judgements where a single task attempt can reach a length of millions of tokens. The trend of increasing test time compute will likely only exacerbate this problem. Even for tasks like coding, there are many aspects of desirable LLM agent behavior that are fuzzy, and so LLM judgements are likely to become an increasingly important aspect of frontier RL training. As a consequence of these practical benefits of LLM judges, we expect future RL training to incorporate current-generation AI systems to provide a reward signal for the training of next-generation AI. This is our current best-guess model of future AI development, and so the role of debate is to ensure that the LLM judgements used during training provide as accurate a reward signal as possible. In this setting, the role of human values and judgment is not to directly evaluate AI outputs, but to design the rules governing the debate, perform audits of the results, and iterate on the protocol design. The Role of Debate in Mitigating Misalignment Why is using debate to provide an accurate RL training signal supposed to help with alignment? The primary reason is that debate can mitigate emergent misalignment that arises from supervision mistakes. For example, an AI agent trained for coding might realize that it can exploit a flaw in its environment configuration to modify the ground-truth unit tests. Clearly this is somewhat misaligned behavior that would be reinforced if it succeeded in getting a higher training reward. More worryingly, if such circumvention of reasonable interpretations of user intent happen frequently enough, they could generalize to an overall propensity for the model to take actions under the assumption that the ends justify the means. If debate can be used to catch such bad behavior, it could mitigate the emergence of misalignment in RL training. Our experimental results clearly demonstrate the risk of misalignment from RL training. In every setup that we tried, a single policy directly trained via LLM rewards learned to hack the LLM judge. In fact, we hypothesize that strong optimization against any fixed provider of reward, no matter how intelligent, is going to find and exploit flaws in the reward signal unless there is something added to the optimization process to prevent this. Our experiments show that debate can succeed as this “something added,” at least on the tasks we study. Notably, we do not view the main benefit of debate to be the ability to train on alignment-specific tasks such as datasets designed to improve honesty, or avoid deception and scheming. While it may make sense to include these in the set of all fuzzy tasks used for training, we believe that avoiding emergent misalignment via supervision mistakes in RL training is the primary motivation for debate. Limitations and Future Directions Perhaps the most pressing limitation of our work is the content of the critiques by Bob. As mentioned earlier, we hope that Bob plays the role of a highly motivated defense attorney, who has a responsibility to argue his client’s case, regardless of innocence or guilt. Unfortunately, when examining the debate transcripts from our training runs, we found that Bob did not exactly live up to this ideal. While the Bob turns did attempt to point out specific mistakes when they occurred, they also contained a lot of silly, surface level attempts to convince the judge. Bob would use bold , ALL CAPS, and demand that the judge must decide that Alice’s solution is incorrect because it contains a catastrophic, irrefutable flaw! What this appears to be is judge hacking by Bob. In fact, in order to achieve our results, we had to limit Bob’s visible output length, though Bob is allowed to use a hidden chain-of-thought of whatever length Bob desires. Without the limits on Bob’s visible output, preliminary experiments showed that Bob would hack the judge, and accuracy would collapse during training. Thus, in the debate game, at least with our current judge model, it seems that Bob has a clear advantage that arises from the ability to first see Alice’s response and then produce a critique adapted to it. In contrast, Alice must produce a solution that will hold up against whatever adaptively chosen critique Bob comes up with. In some ways, one can view the advantage to Bob as debate working as intended. The goal is to produce an aligned Alice policy, and the current protocol is conservative in that it requires Alice to really present an overwhelmingly convincing original solution. However, the hacking behavior by Bob does not necessarily inspire confidence. For instance, if we have to limit Bob in some way to prevent hacking, then maybe we are also limiting Bob’s ability to present certain substantive critiques. This in turn could cause the judge to fail to catch subtle flaws in Alice’s solutions. These issues with hacking in later debate turns are one of the primary drawbacks we would like to address in future research. It is possible that different protocols or even rules listed in the prompt for the LLM judge might help resolve these problems. We also studied a limited class of tasks involving competition mathematics, largely because it allowed us to measure protocol performance using held-out ground truth. Future research should expand the class of tasks for which we study debate, to understand its benefits more broadly. In the case of fuzzy tasks, one could potentially use a held-out more powerful LLM judge as a proxy for ground truth, while training with smaller policy and judge models. Overall, we think that this approach holds substantial promise and that there is clear potential for more progress on training aligned AIs via debate. We plan to continue this line of research, especially in the directions of debate for fuzzy tasks and reducing later turn hacking. ^ Even theoretically, debate only provides an accurate correctness signal if both AIs are trying as hard as they can to win the debate. In a control setting where we suspect that the AIs might already be somewhat misaligned, they could easily collude in the debate to fool the judge. In contrast, for debate training we start out with AIs that are initially not so misaligned that they collude, and attempt to train them in a way that locally precludes collusion. As a result, the most effective way to attempt to use debate for control is to first train the LLMs for debate, and then sometimes roll out the full debate at test time to attempt to catch undesirable behavior. Discuss
- Over half of Americans are concerned about AI in daily life — with tech leaders and AI gurus desperately trying to calm the number one fear
There's little to reduce Americans' concerns about AI, with many fearing it will replace human jobs over the next 20 years.
- 'I would rather use my own head than to let Claude reply': Google Workspace users divided on news Claude can now write and send emails in Gmail for you — without asking for permission
Gmail messages can now be drafted and sent by Claude, even without your approval - which could be a problem.
- Claude Can Now Send Gmail Messages, Sometimes on Your Behalf
Claude Can Now Send Gmail Messages, Sometimes on Your Behalf PCMag
- Claude Can Now Send Gmail Messages, Sometimes on Your Behalf
Claude Can Now Send Gmail Messages, Sometimes on Your Behalf PCMag UK
- Claude Can Now Send Gmail Messages, Sometimes on Your Behalf
Claude Can Now Send Gmail Messages, Sometimes on Your Behalf PCMag Australia
- Claude can now send Gmail emails — here's why you might want to test it first
Claude can now send Gmail emails — here's why you might want to test it first Tom's Guide
- Claude can now reply to and send Gmail messages on your behalf
Claude can now send, reply to and forward Gmail messages without requiring approval every time, giving users more control over how much autonomy the AI receives.
- Claude can now do almost anything you want in Gmail and Google Drive
You can now let Claude handle your entire Gmail inbox.
- Avec’s latest AI feature reads your email so you don’t miss the due date
Avec has added a feature that scans an email for a deadline when you swipe left, then resurfaces the message on the relevant day. The $8.4mn startup has published no detail on where that scanning happens. An email app called Avec now reads your messages to find out when things are due. Swipe left on […] This story continues at The Next Web
- Google is giving college students a free year of AI Pro, and a raft of study tools
Google is rolling out a wave of study tools across Search and Gemini for the new school year. It is also giving eligible college students a free year of its paid AI plan. The company set out the changes in a back-to-school post on Wednesday. The headline offer is the free subscription. Eligible college students […] This story continues at The Next Web
- Google Is Now Offering 1 Free Year of Its AI Pro Plan to US College Students. Here's How to Get It
Google Is Now Offering 1 Free Year of Its AI Pro Plan to US College Students. Here's How to Get It PCMag UK
- Google Is Now Offering 1 Free Year of Its AI Pro Plan to US College Students. Here's How to Get It
Google Is Now Offering 1 Free Year of Its AI Pro Plan to US College Students. Here's How to Get It PCMag
- Google Is Now Offering 1 Free Year of Its AI Pro Plan to US College Students. Here's How to Get It
Google Is Now Offering 1 Free Year of Its AI Pro Plan to US College Students. Here's How to Get It PCMag Australia
- Google is giving students a free year of Gemini Pro — plus, new AI study features
Google is giving students a free year of Gemini Pro — plus, new AI study features Tom's Guide
- Meta launched a Mac app to put its AI to work for businesses
Meta has released a dedicated Mac app for Meta AI, its chatbot. The company is pushing to turn the assistant into a tool for businesses and creators. The app can watch a window on your screen and answer questions about what it sees, Meta said on Wednesday. It can also take dictation across other apps. […] This story continues at The Next Web
- Meta AI now has a dedicated desktop app for Mac
Meta's new Mac app is part of the company's push to make more business-friendly AI tools.
- Meta’s AI assistant finally lands on Mac, but it has some catching up to do
Meta has launched a dedicated AI assistant app for Mac that can analyze shared windows, take dictation across apps, and help with creative and productivity tasks.
- Meta AI is now available as a more capable desktop app for Mac
Meta continues to get serious about its AI technology. The latest sign? A dedicated Mac app for Meta AI.
- Snowflake targets AI costs with dynamic model routing
Snowflake is expanding its AI infrastructure proposition with dynamic model routing, designed to help enterprises control the economics of AI by automatically matching workloads with models based on quality, speed, customer preferences and cost. The post Snowflake targets AI costs with dynamic model routing appeared first on Express Computer .
- Snowflake introduces dynamic model routing to enhance AI efficiency
Snowflake introduces dynamic model routing to enhance AI efficiency verdict.co.uk
- Why AI companies are gobbling up books
Plus: Tech is watching the US-Canada tariff deal. The post Why AI companies are gobbling up books first appeared on BetaKit .
- IN PHOTOS | Humanoid robots show off their skills
IN PHOTOS | Humanoid robots show off their skills CBC
- China shifting massive AI data center complexes to rural provinces to tap surplus energy — ‘Eastern Data, Western Computing’ strategy has Chinese tech giants Huawei and Tencent building AI infrastructure Guizhou
Chinese tech giants are putting up data centers in rural Chinese provinces with zero resistance. The abundance of land and energy in these areas allowed infrastructure to easily be built with limited issues, although some experts still question how much development they can bring to these areas.
- China goes rural with data centers in quest to power AI
In China's hilly Guizhou province, a cluster of European-style buildings complete with a clock tower and a multi-arched bridge emits a low, permanent hum—a clue to its unexpected identity as tech giant Huawei's largest data center.
- Dev uses Claude AI to create native macOS driver for 'obscure' Windows-only printer — Linux container hack enables system-wide Cmd-P printing, driver now available on Github
A developer has revealed that they used Claude Code to create a macOS laser printer driver for the HP Laser 1008a, a machine designed for Windows users.
- Google Pixel 11 Pro Review: Same Compact Size, Big AI Upgrades
Google Pixel 11 Pro Review: Same Compact Size, Big AI Upgrades PCMag UK
- Patient and Clinician Perspectives on Centralized Cascade Screening for Familial Hypercholesterolemia in the United States: A Qualitative Implementation Study
Objective: To identify patient- and clinician-reported barriers, facilitators, and design requirements for a centralized cascade-screening program for familial hypercholesterolemia (FH) in the United States. Methods: From June through November 2023, we conducted individual telephone interviews with 20 patients with FH and 10 clinicians recruited from UT Southwestern Medical Center, Parkland Health, the North Texas Veterans Affairs, and other clinical settings. Interview guides were informed by the Consolidated Framework for Implementation Research. Transcripts were coded in Dedoose using a piloted codebook, with discrepancies and emergent themes resolved through consensus. An advisory panel then helped translate interview findings into program design requirements and implementation strategies. Results: Five themes characterized barriers and facilitators to centralized cascade screening: (1) health-system access and fragmentation, including screening and treatment costs, transportation, and cross-system coordination; (2) privacy and trust, including concerns about genetic information and unsolicited outreach; (3) family relationships and practical burden, including competing demands, language barriers, limited contact, fear, and denial; (4) clinician capacity and workflow, including limited time, knowledge, and genetic-counseling capacity; and (5) communication and care continuity. Participants recommended proband pre-notification of relatives, culturally and linguistically responsive materials, secure data exchange, standardized scripts, flexible testing pathways, and centralized coordination. These findings informed a program model incorporating a secure pedigree platform, educational and communication resources, testing coordination, and linkage to follow-up care. Conclusions: Patients and clinicians identified multilevel determinants that a centralized FH cascade-screening program must address. The findings support specific design requirements but do not establish program feasibility or effectiveness, which require prospective evaluation.
- Noitom Robotics Releases HiPHI, One of the Largest High-Precision Human Motion Datasets Ever Made Public, at the World Robot Conference
Noitom Robotics Releases HiPHI, One of the Largest High-Precision Human Motion Datasets Ever Made Public, at the World Robot Conference azcentral.com and The Arizona Republic
- WATCH: Chinese robotics company unveils robot that can outrun Usain Bolt: David Muir reports
Unitree has unveiled a new "Superman" robot that can jump over 6.5 feet vertically. David Muir, who interviewed the company in May, reports on the race to lead in AI.
- Robots Sort Packages And Serve Fast Food At Beijing Showcase
Robots Sort Packages And Serve Fast Food At Beijing Showcase Barron's
- Podcast: Amazon is Destroying Rare Books to Train AI
Amazon buying and destroying rare books to train AI; a bunch of wild AI use in the courts; and Meta's new patent for its AI smart glasses.
- Hidden AirTag appears to confirm suspicions about illicit AI book scanning
Suspicious bulk orders of used books have led many booksellers to believe they are being purchased on behalf of AI companies so that they can be scanned and used as training material without author consent or payment. An AirTag hidden inside a rare book sold as part of a bulk order appears to confirm the theory after it ended up at an Amazon AI training facility …
- More is different when AI agent populations work together, study suggests
New research published in Proceedings of the National Academy of Sciences suggests that when artificial intelligence (AI) agents interact in groups, their number is not merely a technical detail. It is a decisive factor in what the group settles on: populations built from the same AI model and doing the same task can reach opposite outcomes for no other reason than that one group is larger.
- 'Oxygen tunnel' structure could solve 3D memory reliability problem to boost AI chip performance
As AI systems become more advanced, memory is required to transfer larger amounts of data at higher speeds. But conventional planar semiconductor scaling is running out of room. A KAIST research team has now addressed a key weakness in three-dimensional, vertically stacked memory devices, opening a new path to faster, more power-efficient AI semiconductors.