はち @CurveWeb

IT企業勤務。犬とコーヒーが好き。 HuggingFace → https://t.co/vLGtnH1HMV Note → https://t.co/6X6tuOj7QC LLM, Synthetic data(合成データ), Agent Systemについて発言します Joined March 2021

Tweets

4K
Followers

2K
Following

840
Likes

24K

Ino.Ichi @inoichan

2 weeks ago

tascjさんから関東Kaggler会の発表資料を共有していただきました🙌32BまでのLLMを個人でも届きそうなGPUでフルパラ学習するTipsが詰まってます！かなりちゃんと検証されており取り組み方含めて勉強になると思います！Kagglerに限らず役立つと思うのでぜひ！ #kanto_kaggler speakerdeck.com/tascj/di-4hui-…

0 30 204 12K 138

Elon Musk @elonmusk

2 weeks ago

The @xai Grok 2.5 model, which was our best model last year, is now open source. Grok 3 will be made open source in about 6 months. huggingface.co/xai-org/grok-2

3K 6K 42K 9.8M 5K

Lisan al Gaib @scaling01

2 weeks ago

World Map according to GPT-4o

19 25 748 50K 101

Download Image

はち @CurveWeb

4 weeks ago

Agent Lightning: Train ANY AI Agents with Reinforcement Learning 学習の仕組みとエージェントをなるべく疎結合に扱うことでどのようなモデル、エージェント、フレームワークでもほぼ同様にモデルを学習していけるようにするアーキテクチャの提案（Training-Agent Disaggregation）

0 0 7 735 4

Download Image

はち @CurveWeb

4 weeks ago

解答がはっきりしたタスクへのLLMの性能は人間の手を離れて上がり続けますね。オープンエンドなタスクの性能をどう上げていけるのかが今後のエージェント開発のトレンドになると思います。LLMの推論能力向上だけで対応できるのか、それとも別の仕組みを必要とするのか。

0 0 2 567 1

DailyPapers @HuggingPapers

4 weeks ago

Tencent AI Lab introduces R-Zero! A groundbreaking framework enabling LLMs to self-evolve their reasoning capabilities from zero human-curated data, through an autonomous Challenger-Solver loop.

8 89 514 78K 321

Download Image

はち @CurveWeb

4 weeks ago

GPT-5はSWE-BenchギリギリClaude Opus 4.1超えてて、価格やコンテキスト長を見ても（超長コンテキストが必要な時以外は）開発時のモデルはGPT-5が良いかもしれない。

0 1 5 906 1

Download Image

going to try live-tweeting the GPT-5 livestream. first, GPT-5 in an integrated model, meaning no more model switcher and it decides when it needs to think harder or not. it is very smart, intuitive, and fast. it is available to everyone, including the free tier, w/reasoning!

1K 2K 28K 3.5M 4K

OpenAI @OpenAI

4 weeks ago

GPT-5 is here. Rolling out to everyone starting today. openai.com/gpt-5/

4K 7K 33K 4.6M 4K

Download Video

OpenAI @OpenAI

a month ago

LIVE5TREAM THURSDAY 10AM PT

2K 3K 24K 6.9M 2K

Jiayi Weng @Trinkle23897

a month ago

Harmony format is finally open-sourced. I still remember 3 years ago (before ChatGPT release) @shengjia_zhao, Daniel and I were brainstorming about the right abstraction for RL training, and that is the start point of the entire harmony library. github.com/openai/harmony

34 158 2K 142K 737

はち @CurveWeb

a month ago

blog.langchain.com/deep-agents/

0 0 5 664 2

Daisuke Okanohara / 岡野原大輔 @hillbig

a month ago

階層推論モデル（Hierarchical Reasoning Model; HRM）言語に依存しない階層型の連続空間での思考を実現し、27Mのモデル、千例の訓練データでARC-AGIや数独、迷路などを解けるように学習できる。従来のTransformerは深さが固定かつ思考は人や強化学習で獲得したCoTに頼る必要があった。…

1 54 286 37K 213

Mark Kretschmann @mark_k

a month ago

OpenAI open source model FOUND! Name: gpt-oss-120b Size: 120b Release date: TODAY! @OpenAI

50 92 1K 119K 230

Download Image

Aryo Pradipta Gema @aryopg

2 months ago

New Anthropic Research: “Inverse Scaling in Test-Time Compute” We found cases where longer reasoning leads to lower accuracy. Our findings suggest that naïve scaling of test-time compute may inadvertently reinforce problematic reasoning patterns. 🧵

59 175 1K 323K 634

Download Image

Jackson Atkins @JacksonAtkinsX

a month ago

LLMs can now self-optimize. A new method allows an AI to rewrite its own prompts to achieve up to 35x greater efficiency, outperforming both Reinforcement Learning and Fine-Tuning for complex reasoning. UC Berkeley, Stanford, and Databricks introduce a new method called GEPA…

48 337 2K 204K 2K

Download Image

fly51fly @fly51fly

a month ago

[LG] Does visualization help AI understand data? V R. Li, J Sun, M Wattenberg [Harvard University] (2025) arxiv.org/abs/2507.18022

0 18 56 75K 27

Download Image

elvis @omarsar0

a month ago

Deep Research Agents with Test-Time Diffusion Google keeps pushing on diffusion. This time, they apply diffusion to deep research agents, specifically the report generation process. It achieves a 69.1% win rate vs. OpenAI Deep Research on long-form research. My notes: