@OfficialLoganK "let's keep shipping minor models that nobody is asking for when our major models are still largely behind in coding."
English
Elliot
562 posts
Introducing Nano Banana 2 Lite 🍌 and Gemini Omni Flash 🔮, our new generative media models in the Gemini API and AI Studio!
Nano Banana 2 Lite is extremely fast (<4s image) & cheap ($0.034 / 1K image).
Omni Flash is SOTA at video editing at $0.10 / sec, same as Veo 3.1 Fast!
English
Google 即将发布Nano banana 2 Lite和Gemini Omni Flash🔥🔥🔥
no Gemini 3.5 Pro😭
Logan Kilpatrick@OfficialLoganK
⚡️🍌🔮
Eesti
@testingcatalog I understand the purpose of flash model,but flash lite is straight trolling
English
GOOGLE 🔥: A new Nano Banana 2 Lite image generation model is on the horizon!
> "Introducing Nano Banana 2 Lite for ultra-fast, cost-efficient image generation, and Gemini Omni Flash for powerful video creation with conversational, step-by-step editing"
Logan Kilpatrick@OfficialLoganK
⚡️🍌🔮
English
冷知识：MIT/Stanford为了保证自己的逼格，都不会发正经理工科数理化生计的master给外人，我了解到一般只有本校直研和PhD Quit才有master——所以我们知道这是一个镀金水硕。
疯越飞人院@runawayfromtheh
我刚看见一个读了MIT硕士的说自己原生家庭差到离谱，就因为家里亲戚有一半不知道MIT这个学校😅
中文
@Hadas_Gold @theinformation Yes, please go slow and let China take the rest of the world.
English
I've confirmed this from @theinformation. OpenAI is doing this as they recognize it's essentially the only way they can then release GPT 5.6 to the general public. It's very confusing who/what is 'regulating' new AI models in Washington.
Stephanie Palazzolo ✈️ ICML@steph_palazzolo
New w/ @leomschwartz @amir: The Trump admin has asked OpenAI to stagger the release of GPT-5.6 over security concerns. On Thursday, CEO Sam Altman told staff that the government will be approving access to GPT-5.6 customer by customer, a highly unusual approach.
English
我来给我自己招 leader 了，大家有想来苏州发展的大佬们，欢迎投递，要求见图片。
美股上市公司 GigaCloud Technology Inc(NASDAQ: GCT)
中文
如何加入OpenAI？只需精通以下课程：
1. 斯坦福大学的“从零开始的语言建模”课程：cs336.stanford.edu/spring2025/
2. 在掌握广度之后，她逐一深入研究概念，使用博客、论文、与 ChatGPT 和 Claude 聊天，以及从零实现相关内容。
3. 在面试中，实现/调试 Transformer 的问题经常出现。将它转化为肌肉记忆：github.com/stanford-cs336…
4. 刷 Leetcode🥲 leetcode.com/studyplan/leet…
5. 她分享的其他学习资源：
a. 自注意力 & Transformer：web.stanford.edu/class/cs224n/r…
b. 插图版 GPT-2：jalammar.github.io/illustrated-gp…
c. 反向传播
cs231n.github.io/optimization-2/
d. 语言模型策略梯度入门
ivison.id.au/2026/02/09/pol…
e. 理解 GRPO 和强化学习原理的轻量级指南
gitlostmurali.com/blog/grpo-intr…
f. 如何扩展你的模型
jax-ml.github.io/scaling-book/
这是我能免费看的内容吗？🤡
Alisa Liu@alisawuffles
I'm joining OpenAI next week!🥹 The job search turned out to be really challenging but also super rewarding, so I wrote a small blog to share what I learned along the way and hopefully make the process a little less mysterious for the next person. alisawuffles.github.io/blog/job-search
中文
谷歌 TPU 软件工程师 Patrick Toulme 指出，外界对 GLM 5.2 靠蒸馏追平 Opus 的说法存在误解。大模型在智能体编码任务上的训练难点在于「零梯度困境」，即模型早期若无法产生正确运行路径，强化学习便无法获得梯度信号来启动参数更新。蒸馏 Claude 或 GPT-5.5 的作用，仅仅是在冷启动阶段提供种子解答以绕过零梯度困境。
一旦模型跨过冷启动门槛，后续的性能爬升将不再依赖蒸馏，而是完全依靠强化学习的爬山算法进行自我演化。Toulme 强调，GLM 5.2 已经具备独立产生成功路径的能力，完全可以通过强化学习自主迭代到更高级别，彻底摆脱对美国大模型的依赖。
Patrick C Toulme@PatrickToulme
There’s a big misconception about how GLM 5.2 was trained. Yes, they distilled Claude and GPT 5.5 — but distillation is not how they matched Opus quality. Distillation only fixed the cold start problem in RL. RLing an agentic coding model isn’t rocket science. In simplified terms: 1. RL needs trajectories — rollouts where the model actually completed a task in some env 2. No successful trajectory on a task = zero gradient = you can’t RL it. This is the cold start problem 3. Distillation solves it. You seed your model with knowledge from a smarter one (Claude, GPT) on tasks it can’t do yet 4. Now it produces positive trajectories on those tasks 5. RL on those trajectories and hill climb agentic coding 6. At that point you no longer need to distill and can solely hill climb RL to better models This is an interesting curve. I’d argue it’s harder to get to Opus 4.8 from scratch than to go from Opus 4.8 → Fable/Mythos tier. GLM 5.2 is already producing positive trajectories, so they have plenty to RL on — they’ll keep climbing to Mythos quality without distilling any further. They no longer need American models.
中文
I just talked with someone from Google DeepMind tonight
they said that all the ppl who recently left (e.g. Noam Shazeer) were working on some super-hard-to-land stuff by themselves, and since they were working alone anyway, their departure does not affect Google DM that much
the rest of the DM ppl were actually pretty pumped and working super hard to ship the next model which will be out soon and will have a big leap in performance
jinjingliang@JinjingLiang
Google is getting left behind. With this and the departure of some of their top talent this week... 100% unexpected, but still better than Meta 😂
English
I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there.
It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to work with all of you.
English
@PolymarketMoney @claudeai By the end of the year fable will be outdated bruh
English
@claudeai There's now just a 50% chance Claude provides its U.S customers with access to its new "Fable" model again by the end of this year.
polymarket.com/event/claude-f…
English
