All AI news

AIWe have to batch our thoughts. We can’t point at things. We phrase questions like emails. The interface doesn't leave room for us so we adapt to the models.

AIAndrej Karpathy says asking an LLM to structure its response as HTML and viewing it in a browser works well, and that slideshows have also worked for him. He argues vision is the preferred AI output channel, outlining a progression from raw text and markdown toward HTML and eventually interactive neural videos, while input methods like pointing and gesturing still need improvement.

AIIt's a real privilege to work with people so deeply motivated to make the future go well!

AIJan Leike says alignment research has grown from a dozen side-gig researchers into a field the world increasingly recognizes as important. He credits RLHF on LLMs with making alignment more practical, along with progress on evaluating and fixing behavioral issues. He also notes Claude now has a constitution and that more alignment research is being automated.

AI…and the stakes keep getting higher.

AIBerkeley AI Research describes adaptive parallel reasoning, in which a reasoning model decides when to split independent subtasks, how many concurrent threads to spawn, and how to coordinate them. The approach targets the latency, context-rot, and cost problems of long sequential reasoning, which can require millions of tokens and tens of minutes for complex tasks. Existing methods such as self-consistency, Tree of Thoughts, ParaThinker, and Hogwild! Inference fix the parallel structure outside the model, which wastes compute on simple problems.

AIMETR's external review of the "Risks from automated R&D" section of Anthropic's February 2026 Risk Report concludes the report does not adequately support its claim that catastrophic risk from Claude Opus 4.6 or a less capable model automating R&D is very low.

AIWriter Dean Ball says he opposes nearly all proposed AI regulation, including algorithmic discrimination rules and pauses on development, while backing state management of catastrophic misuse risks. He frames the position as a tension between classical liberal and conservative instincts toward institutions and change.

AIDean W. Ball argues that Anthropic's Mythos, which finds software vulnerabilities by chaining bugs into exploits, shifts the cost of vulnerability discovery and should not prompt an overreaction. He contends governments hold a uniquely mixed incentive over vulnerabilities, so heavy state control risks making software less secure. He proposes a narrow, testable government role focused on cyber-discovery risk thresholds, with private verification bodies supporting it.

AIAndrej Karpathy argues that LLMs enable new functionality, not just faster versions of existing software, citing an image-to-image app, markdown skill files replacing bash install scripts, and LLM knowledge bases over unstructured data.

AIAndrej Karpathy describes a December 2025 shift in which coding agents began producing larger, more reliable chunks of work, changing programming toward orchestrating agents. He argues that models automate what can be verified and that their capability is jagged, depending on verifiability and what labs emphasize in training, so users need to stay in the loop. He also says hiring, founder opportunities, and agent-native infrastructure should adapt to this shift.

AIBiological data is the missing link. It may not be sexy or make for shiny announcements but building biological infrastructure is where the impact is. Huge props to CZI.

AIAnthro probably needs to scale up Account Support via Claude or via humans (traditional account managers). Would be funny if they chose humans. Alternatively, more and more companies can probably go multi-AI with open harnesses. Similar problems to the cloud era enterprise pains. I'm sure these issues would apply to all the other AI providers.

AIMckay Wrigley says his coding split moved from 80/20 Claude/GPT to 80/20 GPT/Claude within three months, and he trusts GPT 5.5 for engineering. He still finds Claude better for non-coding agent work, calls Opus 4.7 underwhelming, and attributes Anthropic's issues to compute constraints.

AIAhmad Al-Dahle argues that the most interesting part of DeepSeek-V4 is its bet on efficient ultra-long context rather than its benchmarks. He says this is the precondition for test-time scaling and long-horizon agents, and cites 27% of V3's FLOPs at 1M tokens. The quoted DeepSeek post announces DeepSeek-V4-Pro (1.6T total, 49B active) and DeepSeek-V4-Flash (284B total, 13B active), both open-sourced with 1M context and API access.

Why it matters: The post argues that efficient 1M-token context, not benchmark scores, is the key bet behind DeepSeek-V4's design for test-time scaling and long-horizon agents.

AICognition argues that enterprises building cloud agents face three problems: shared container kernels, the inability to persist agent state across async gaps, and the scale of orchestration, governance, and integrations. The post says VM-level isolation with hypervisor-level snapshots was needed for Devin, and that organizations must also rebuild engineering processes around agent execution.

AIRavi Nagayach, Prashant Singh, Kshitij Gupta, and the Lambda networking team have been doing exactly that for about a decade. Read their story here:

AIThey have to be extreme co-design companies (like Nvidia, but even more). That’s where the global minima are. I’m especially thinking about Apple and how with the recent transition a lot of people are saying it’s a hardware company. It should be (and hopefully will be) extreme co-design.

AIEugene Yan shares a Mozilla writeup reporting that Mythos found 271 vulnerabilities fixed in Firefox 150, while Opus 4.6 found 22 fixed in Firefox 148. Mozilla quotes its finding that no category or complexity of vulnerability humans can find has been beyond the model so far.

AISoumith Chintala says Jensen Huang understood AI ecosystems, trade, and policy far better than host Dwarkesh Patel in their podcast. He argues that no single model such as Mythos marks a critical phase change, since a state-of-the-art Chinese open-source model with three orders of magnitude more test-time compute and unpublished post-training advances would be a more realistic baseline. He also says American policy should use measured, continuous levers across a Western-controlled ecosystem rather than abrupt interventions.

AIJason Wei describes a trend where individuals gather continuous health data, such as lab tests and wearable metrics, and use AI to understand their bodies. He says he moved from six-month to six-week Function Health blood tests to see how lifestyle changes affect him, and uses AI to analyze the results.

AIThis is why we chose to point the AAR at this scalable oversight problem! Progress would let AARs work on fuzzier alignment problems, where humans can only provide weak supervision.

AIwhere we can procedurally verify the quality of the work: automated red teaming, auditing games, control methods, etc.

AI…trained on “golden” data can be recovered when training the strong model only using a weak supervision signal? This blog post explains it:

AIone skipped the weak teacher entirely after noticing the most common answer was usually right. We caught these, but it's a warning for future AARs whose hacks may be harder to catch.

AI…specific subproblems for AI to solve). Most of our work is the former (e.g. using Claude for coding), but this work is about the latter.

AIAndrej Karpathy argues that people judging AI from free-tier ChatGPT or Advanced Voice Mode miss the strong capabilities of current agentic models like OpenAI Codex and Claude Code. He says gains are "peaky," concentrated in verifiable technical domains like programming and math that suit reinforcement learning and attract B2B investment, while writing and everyday advice improve less. Those who use frontier agentic tools professionally in these fields see far greater capability, which is why the two groups talk past each other.

AIThe final Glasswing model is less likely to do things like leak information, though it's still somewhat pushy, and at least as capable of doing things like working around sandboxes.

AII'm excited about much new research we have feeding into the welfare assessment for this model: