@sun_hanchi @antirez 50b and 18b are valuation of latest funding round, not closed yet
English
Shuyao Tim Xu
64 posts
Will GPT-5.5 surpass top1?
"16MB artifact size" will induce some parameter efficient but not inference efficient methods.
This is also interesting, cuz it targets Kolmogorov complexity.
(I'm not very interested in this setting tho)
OpenAI@OpenAI
Are you up for a challenge? openai.com/parameter-golf
English
@YouJiacheng @CyouSakura stepfun reasoning team is indeed very strong
English
Step 3.5 Flash is surprisingly strong.
cc @CyouSakura
Jasper Dekoninck@j_dekoninck
The results are not only poor, differences between models are also huge: Opus-4.6 completely fails, and Gemini-3.1 performs only half as good as GPT-5.4. The capability does not seem stressed equally among all frontier models, even though it is essential for math use cases.
English
best perk of working at an LLM shop:
your agent system is blocked on infra, so you get to go home like a normal human being for once
English
比较靠谱的猜测，hunter是蚂蚁的Ling-2.6-1T。
healer是小米的模型。
向阳乔木@vista8
求证：OpenRouter新上的两个隐身模型是DeepSeek V4吗？ 发现自 @geekbb 一个叫Healer Alpha（治疗者Alpha） 具有视觉、听觉、推理和行动能力的前沿全模态模型。 原生感知视觉和音频输入、跨模态推理以及精确可靠地执行复杂的多步骤任务。 一个叫Hunter Alpha（狩猎者Alpha） Hunter Alpha是为Agent使用构建的1万亿参数+1M Token模型。 擅长长期规划、复杂推理和多步任务执行。 具有OpenClaw等框架所需的可靠性和instruction-following精度。 感觉这个AI团队估计是喜欢MMORPG游戏的...
中文
@dizhyun510754 @teortaxesTex nah deepseek has a unique style of response to those queries
English
@TimXu222575 @teortaxesTex All of the Chinese models cant talk about political sensitive, it dosen't make sense
English
If Hunter-Alpha is V4 and it really has 1T tokens, that's very sad (unless it was trained on Ascends, which would suggest better models not constrained to 2-4K H800s to come soon. But still sad on the V4 side)
If it's something like Kimi-DSA/Xiaomi/GLM, that's better.
English
@scaling01 kimi agent (the feature on web) has it alalready. One advantage is that you create a full website from scratch with ai generated figures. However i agree this is more for cool demo
English
I don't know who needs this but Codex CLI will support image gen natively in the next big update
English
@scaling01 gpt oss was actually a great model. It has the proprietary that chinese model lacks
English
For some reason I dreamt that OpenAI is going to release two open-source models today
but I guess GPT-5.4 will also do
English
what the heck? hr was scheduling my interview with GOAT @JustinLin610 just last friday🤯🤯🤯
English
@Teknium Including DeepSeek is purely political here. It feels like DeepSeek is just using Claude for comparison, maybe comparing their judge with Claude's.
English
FYI 150k rubric judgements is just 5000 sample RL training run at groupsize 32. This is like, literally nothing.
Fear mongering. DeepSeek gets such a bad rap because Anthropic and our labs are complete shitbags
English
Ohhh nooo not my private IP how dare someone use that to train an AI model, only Anthropic has the right to use everyone elses IP nooooo, this cannot stand!
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
@pmddomingos Did you notice that the DeepSeek's requests is only at the scale of 150,000??
Including DeepSeek in the report seems purely political. It is nowhere "distilling"
English
DeepSeek, just another great Chinese success story.
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
These attacks are growing in intensity and sophistication. Addressing them will require rapid, coordinated action among industry players, policymakers, and the broader AI community.
Read more: anthropic.com/news/detecting…
English
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax.
These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
@Ryan76589177 @caolei1 1. 豆包线上请求最多的肯定是flash模型，而不是pro模型
2. 单卡A100，不做任何多卡优化，跑gpt-oss 20b可以上万tps
中文
@jayair K2.5 thinks very efficiently and is much faster then recent opus 4.6 or codex 5.3. It reminds me of the old sonnet (speed wise). K2.5 is a compelling daily driver.
English
