Joined August 2023
Watched Gemini at Work 2026. Has Google now become an Indian company? The management team is all Indian!
11
Tibo said it’s faster, so I ran a rough test: Standard: 43.6 tokens/s Fast: 56.0 tokens/s Standard is faster now, but GPT-6.1 Sol Fast costs 2× as much for just 28.5% more speed. Double the price for 1.29× the speed? That feels like poor value. Why so little gain?
19
Openai 实在是太精了!GPT-6.1 Sol 便宜,是通过砍速度得来的。 开 fast 模式吧,到头来还是得加钱😅。 那价格跟GPT-5.6 Sol 一样了,但真正的大头缓存输入还是便宜一半。
19
codex底层可以调用吗?这将大幅提高任务完成速度。或者说以后大模型训练都应该有两个出口,一个面向用户的文字说明,一个面向代理应用的快速决策。
Give your app real-time decision-making with Decisions API, powered by GPT-6 Luna. Define questions and possible answers to classify content, route requests, or choose an agent’s next action. Available in limited preview.
30
OpenAI算力短缺,所以他们在故意让用户退订。😅
9
Serious bug: the ChatGPT Windows app became completely unlaunchable after an update. The new version downloaded successfully, but AppX registration repeatedly failed with 0x80070005 / 0x80073CF6 and left the package stuck in Servicing. I could only recover the previous version.
1
1
57
Windows版ChatGPT更新后出现0x80070005 是更新被拒绝 在 PowerShell $pkg = Get-AppxPackage -Name OpenAI.Codex $manifest = Join-Path $pkg.InstallLocation 'AppxManifest.xml' Test-Path $MANIFEST 应返:True 再输: Add-AppxPackage -DisableDevelopmentMode -Register $manifest
1
34
ML945 retweeted
gpt 6 sol is actually terra been using sol since it came out and man something feels off it doesn’t feel like sol at all, it feels way more like terra. even price is almost same as 5.6 terra i think openai upgraded terra, changed name and gave us new model could be completely wrong, but if sol was distilled from astra, shouldn’t it be closer to astra? either this is upgraded terra or some kind of sol minor, and we might see real sol upgrade at dev day am i tripping or does anyone else feel this too?
OpenAi is preparing Astra Minor model just appeared in official docs alongside astra, sol and luna it’s listed for daybreak access too, with reduced refusals for approved red users but there’s still no model page, pricing or benchmarks so i think it’ll be cheaper version of astra, maybe even better at some tasks kek don't think team is gonna release it this week so we gonna have wait till dev day now
43
15
1
395
30,930
MiMo-V2.6-Pro 这是屠榜了吗? 开源第一情况下比Luna还便宜😱
37
闭源才是真正的危险。 由一群满嘴仁义道德的虚伪骗子来控制最强大的闭源AI,其他人没有反击的机会,沦为鱼肉,人类才会毁灭。
30
AI早晚会强大到灭亡人类! 不如我控制最强大AI ,监控全人类,成为掌控人类灭亡按钮的统治者!!! 你们的反乌托邦算盘打的真响。
We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them. We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies. These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve. We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop. Read the report: anthropic.com/threat-intelli…
58
由于`gpt-6-astra` 过于昂贵 以及研究发现:额度消耗的真正占大头是父代理每次工具往返都要重新带入的上下文。 所以我以后主要改为父代理用 Sol,gpt-6-astra作为可以调用的子代理专家。
65
我的Codex子代理调用与精简流程插件 已升级到 3.2.0。 这次改变了父代理的路由行为: 防止昂贵父代理把一次“直接处理”决定沿用到整个长任务。 把父代理继续独立工作的 token、时间和返工也计入路线总成本。 遇到已经暴露出来、适合独立处理的困难时,及时调用有针对性的专家,而不是由父代理长时间试探
1
110
codex时间策略并未更改,使用银行重置后,继续随重置顺延。以后尽量少用GPT-6 Astra,真的用不起。
37
ai是新的生产力革命
连训模型的人,都快被模型创飞了。 刚看朋友分享了一个大佬的朋友圈,挺震撼的。 他让 GPT-6 去复现自己大一时候做的一篇电路布线论文。 这不是随便一个课设。当年这东西发了 CCF A 类期刊,他还申请过专利。 结果 GPT-6 一分钟就复现完了。 然后他又临时想了一个改进方向,让模型继续往下做,不到半天,新的方案也弄出来了。 这件事最冲击我的其实不是一分钟浮现论文本身。 而是发这条朋友圈的人,本身就是参与训练 GPT-6 的人。 也就是说,连最接近模型、最清楚它能力边界、甚至亲手参与把它训练出来的人,真正开始拿它做自己的研究时,还是会被它吓到。 他后面有句话我特别有感触。 以前做 research,很像一次特别长的 RL rollout。一个方向可能要做几个月甚至几年,最后才知道对不对。 但现在,一个想法上午提出来,下午可能就已经验证完了。 所以他说,以前人总幻想“朝闻道,夕死可矣”。 但如果以后真的可以天天朝闻道,每天都有新的答案、新的方法摆在你面前,那种感觉其实很难形容。 可能很多人现在还觉得,GPT-6 不就是又一个更强的模型。 但我这两天越来越明显地感觉到,真正做 AI、做研究、做工程的人,对GPT-6反应其实都差不多。 大家不是没想过这一天会来。 只是没想到真来的时候,还是会这么震撼。 更没想到,会这么快。 PS:这个佬挺活跃的,大部分人都知道吧?
17
现在肯定有人想提取完整的人类🧠神经元。
the fly brain can play beat saber
1
36
靠,我花费大量订阅额度来改Codex 插件。 GPT 6 太贵了,玩不起了。
20