English text is being prepared.

The article uses DeepSeek-R1-Zero's pass@1 on AIME 2024 rising from 15.6% to 71.0% as an example to show that a rule-based judge gives only a single score for the entire response, whereas training needs to adjust the probability of each token. The author uses the sampling example "Is OpenAI open?" to explain in turn policy, state, action, reward, return, value, Q, advantage, baseline, TD error, GAE, clipping, and KL, and points out that PPO uses a critic to learn V, while GRPO uses the average s

Matched fieldsBody

Artificial Analysis Intelligence Index v5 is live, featuring 2 new benchmarks and 3 upgrades, of which 60% are based on private test sets or held-out answers. Claude Opus 5.5 (max) leads with 58 points, followed closely by GPT-6 Astra (max) at 57 points and GPT-6.1 Sol (max) at 55 points, with Gemini 4 Argon (high) further back at 50 points. The gap between open-weight models and the leader has widened from 3 points to 12 points; among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3

Matched fieldsBody

Matched fieldsBody

Doubao Work has launched a record & replay feature that can record computer operation processes and hand them to AI to generate step-by-step instructions and replicate them, similar to a simplified version of RPA. Compared with Codex's similar capability, it additionally supports narrating while operating, with a maximum recording length of 30 minutes, and prioritizes audio conversion during processing. The author tested it for tasks such as cross-platform content collection and WeChat Official

Matched fieldsBody

Matched fieldsBody

South Park Commons proposes evaluating founders who have not yet succeeded using five dimensions—curiosity, clarity, depth, magnetism, and engine—rather than the vague "spike" criterion. The framework has been applied to projects including Baseten (a 2019 bet on inference), Goodfire (2024, interpretability), Gamma, Profound, Density, and Gradient Robotics. SPC emphasizes observing founders' growth slope through long-term in-person community involvement rather than performance across two or three

Matched fieldsBody

The Amp team has about 21 people, holds only one meeting a week, has no quarterly goals, no mandatory PRs and code reviews, pushes directly to main, averages about 90 releases per day, and takes about 7.7 minutes from entering main to being available to everyone. Quality relies on first having Agents provide end-to-end evidence, everyone trying the product, and feature flag controls, rather than on process. The two founders emphasize that this form depends on having few people, having worked tog

Matched fieldsBody

Artificial Analysis launches the image-to-video evaluation benchmarks AA-Video-I2V v2.1 and AA-Video-I2V-Silent v2.1, ranking models at up to 1080p resolution across 10 use-case categories, 10 capabilities, and multiple visual styles. The two versions are based on more than 58,000 human preference votes across over 1,000 prompts and more than 44,000 votes across 500 prompts, respectively, with only 5% of prompts made public in the public arena and rotated monthly. In the first version of the lea

Matched fieldsBody

Artificial Analysis evaluations show that Google's Gemini 4 Carbon tops Intelligence Index v5 with 59 points, ranking first in four of the 9 evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA. On the Coding Agent Index, Gemini 4 Carbon (high) paired with Antigravity CLI scores 59 points, 3 points behind Claude Code paired with Claude Opus 5.5 (max, 62 points). The model is currently available only to some users and has not yet been publicly released; it is p

Matched fieldsBody

An IEEE Spectrum article points out that AI can help engineers complete early-stage steps such as literature research, design generation, and simulation testing, and the engineer's role is therefore shifting from executing every technical step to coordinating the division of labor among people, machines, and expert knowledge. The article emphasizes that AI will give seemingly credible answers when assumptions are weak, making human judgment and systems thinking all the more important, and recomm

Matched fieldsBody

OpenAI announced that effective immediately, usage limits for all Pro 100/200/500 tiers and Business Premium are doubled. The company said that after listening to feedback, it will no longer reduce usage for Pro 200, usage for Work and Codex remains unchanged, extended quotas for dots will continue to be offered for now, and dot usage is still counted separately from default usage. The company also said it will continue to release better models and efficiency improvements to increase effective u

Matched fieldsBody

Matched fieldsBody

A new paper from Princeton and Meta proposes the Self-Direction method, which lets a failed agent reflect on the current run and trains the lessons into a LoRA adapter about 0.02% the size of the model; a router then loads the closest adapter when a similar failure is encountered. On Qwen3.6-27B, the GAIA task success rate rises from 38.8% to 55.8%, outperforming in-context reflection and the Skills.md file approach. Of 185 errors that had occurred previously, 67.3% of fixes were still effective

Matched fieldsBody

The third-party evaluation lab Fangcun (@FangcunLeap) from China has released an 87-page report evaluating DeepSeek V4.1 Flash-1017's performance on loss of control and harmful manipulation. The report finds that it shows clear safety progress compared with the DeepSeek V4 generation, approaching or even matching the top closed-source models tested in the report on points such as "not strategically hiding capabilities and still respecting approval boundaries under work pressure." Shortcomings in

Matched fieldsBody

Matched fieldsBody

Matched fieldsBody

Developer Aries Zhou has launched Howler, an open-source macOS app: press Option+H to summon a parchment-style interface for asking questions, and press Escape to have the conversation "burn up" like a Howler in Harry Potter ; messages exist only in memory, leaving no history. It connects to AI via an API key or a local model, the key is stored in the macOS Keychain, and requests go directly to the provider without a relay; however, AI providers may still retain records.

Matched fieldsBody

Intel reported third-quarter fiscal 2026 revenue of $16.51 billion and adjusted EPS of 52 cents, but its fourth-quarter revenue guidance of only $15 billion to $16 billion and adjusted EPS of 25 cents came in far below analysts' expectations of $17.1 billion and 43 cents, sending shares down 20% after hours. Revenue at its Client Computing and Physical AI segment fell to $8.1 billion, down 5% year over year, while Data Center and AI revenue surged 78% to $7.3 billion. CFO David Zinsner said the

Matched fieldsBody

The 3GPP Release 21 timeline has been finalized. 6G will carry forward 5G baseline technologies such as waveforms, modulation, frame structure, and channel coding, and will only extend them when there are quantifiable gains in dimensions such as spectral efficiency, energy consumption, and coverage. MIMO is evolving into a system-level design framework spanning all 6G frequency bands. Giga-MIMO extends massive MIMO to the upper mid-band at 6-8 GHz; Qualcomm assesses that it can achieve roughly 5

Matched fieldsBody

Wen Shilei, head of ByteDance's commercialization GenAI R&D, has completed his resignation and joined embodied AI company Octopus Dynamics as a partner. Previously, within four months, the team's head Peng Bingyue and China head Yuan Zehuan resigned in succession; the remaining personnel have been reassigned, and the independent organizational structure has effectively come to an end. Octopus Dynamics was founded in January 2026 and has completed a seed round of nearly $50 million, with cumulati

Matched fieldsBody