All AI news

AIEpoch AI reports that AI models instructed to improve on several benchmarks could already be boosted by a recent human-authored post-training method, on-policy self-distillation (SDPO). The post implies these gains were known before the models' own technique was evaluated.

AIEpoch AI gave Fable 5 and GPT-5.6 Sol 3000 GPU-hours each to develop a new post-training technique intended to outperform the standard GRPO baseline. The post reports the task setup but does not state the results.

AIAI systems already automate many individual tasks in AI research and development, according to Epoch AI. Full automation of AI R&D, however, would require AI to independently generate and investigate research ideas.

AINVIDIA announced a broadcast covering what's new in CUDA 13.4, with the post linking to a video presentation. The post provides no specific features, figures, or details beyond the title.

AIOpenAI is rolling out GPT-6 with Intelligent UI globally to Plus, Pro, Business, and Enterprise users today, with Free and Go users following starting tomorrow. Plus, Pro, Business, and Enterprise tiers are powered by GPT-6 Sol, while Free and Go tiers use GPT-6 Luna, and both are tuned for everyday conversation. The update applies only to the Chat tab, and the models powering Work and Codex are not changing.

AIAnthropic introduced Claude Haiku 5.5, which it calls the cheapest, fastest, and most capable small model it has released. On average, it costs about 75% less to run than Claude Haiku 4.5. The post positions Haiku 5.5 for high-volume work alongside Opus handling heavier reasoning tasks.

AIAnthropic has introduced Claude Haiku 5.5, which it describes as the cheapest, fastest, and most capable small model it has released. On average, running it costs around 75% less than Claude Haiku 4.5. The author's own post only announces availability and offers no further detail.

AIPreference Model is open-sourcing Karotte, the framework it uses to build reinforcement learning environments that resist reward hacking, including defenses like killing stray processes before grading and rejecting grader-crashing files. The framework has been hardened through more than a million evaluation runs and controlled red-teaming. The company focuses on machine learning engineering tasks for leading labs, and a16z says it is partnering with Preference Model and its founders, Jennifer Zhou and Ning Cao.

AIAnthropic has introduced Claude Haiku 5.5, which it calls the cheapest, fastest, and most capable small model it has released. On average, running it costs around 75% less than Claude Haiku 4.5.

AIAnthropic is halving the price of cache reads on Claude Sonnet 5.5 to $0.10 per million tokens. The company says this makes Sonnet 5.5 roughly 20% cheaper to run on most long-running work.

AIAnthropic has released Claude Haiku 5.5, available now across all platforms including Amazon Web Services, Google Cloud, and Microsoft Azure. The post links to Anthropic's announcement page for further details.

AIAnthropic says Claude Haiku 5.5 shows major improvements across almost all of its alignment evaluations compared with Claude Haiku 4.5. The post reports far fewer instances of misaligned behavior in the new model.

AIAnthropic's Haiku 5.5 is built for high-volume, cost-sensitive work such as summaries and classification. It can serve as a subagent alongside Claude Opus 5.5 and Sonnet 5.5 on coding tasks. It is also fast enough for live customer support and browser use.

AINVIDIA's developer account promoted a broadcast covering what's new in CUDA 13.4. The post provides only the release name and a link, so no specific features, benchmarks, or details are stated.

AIOpenAI unveiled an Intelligent UI update for ChatGPT that generates custom visual elements when they help answer a question. The update is powered by GPT-6, rolling out to paid users immediately and to free users the next day. A reviewer's test produced an annotated slug diagram, an apartment affordability calculator with sliders, and a clickable airplane seat explorer, and OpenAI says users can ask for fewer visual outputs.

AILucas Beyer says robotics is accelerating in the physical world, not just in AI research. He attributes this partly, though not only, to progress in coding models over the past year. The post cites a related thread on scaling a UMI data collection operation from 5 to 90 operators and reaching over 1M unique tasks.

AIMichael Smith was sentenced to 18 months in prison and ordered to forfeit $8,091,843.64 for a streaming fraud scheme that used 10,000 bots and AI-generated songs to inflate streams. The U.S. Department of Justice argued the scheme cut into the royalty pool shared by genuine artists, reducing payouts across the board. Smith's lawyers had sought probation, arguing the case was an example being made of him.

AINvidia announced DGX Station for Windows, a desktop AI supercomputer built on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip with up to 748GB of unified memory, able to run models of up to about one trillion parameters locally. The machine offers up to 20 PFLOPS of AI compute and combines 252GB of HBM3e GPU memory with 496GB of LPDDR5X CPU memory. It is scheduled to go on sale in the fourth quarter of 2026.

AIVercel says Glyph Cluster, listed as stealth/glyph-cluster, is now free on AI Gateway for a limited time while in stealth. Access is restricted to paid users, the model is positioned for coding and agentic knowledge work, and prompts may be used for model improvement.

AIllama.cpp can distribute inference across heterogeneous devices through the ggml RPC backend, according to Georgi Gerganov. He says it is currently an advanced setting, but he expects it to become more accessible to regular users over time. A related post reports MiMo 2.6 Flash running across an RTX 6000 GPU and an M5 laptop over 10 GbE at about 40 tokens/sec.

AIGitHub reports that one in three pull requests now involves an AI agent, and that public secret exposures rise with the volume of pushes rather than from declining developer care. It introduces a ModernBERT-based classifier with Microsoft Applied Sciences that evaluates candidate secrets in under two milliseconds and could more than double the secrets prevented at push time. The feature is in private preview, with availability for GitHub Secret Protection customers later this month.

AILiquid AI and NVIDIA Robotics announced a collaboration around Liquid AI's models on NVIDIA hardware. The main post is brief and gives no further technical detail or figures.

AIDongxi NLP recommends Sherpa, a framework for training LLMs to teach adaptively, tailoring instruction to individual learners rather than simply solving problems. The post presents this as a way for AI to support human learning instead of replacing human teachers, citing the principle of teaching according to each student's aptitude.

AIUnsloth says users can train their own Decision Model on as little as 2.5GB of VRAM using small models such as Laya. Training is done through a UI, with a video tutorial and guide linked from the post and the project's GitHub repository.

AIGary Marcus argues that OpenAI's math announcement omits the procedure, the model architecture, and the failure rate, so its generalizability cannot be assessed. He says it could be a step toward AGI or a Lean-based verification trick in a verifiable domain, and the initial report cannot distinguish the two. The post includes a quoted Terence Tao post that shares a satirical press release about a fictional film-endings repository.