开启时会模糊预览图，关闭后正常显示
![About using LLMs to help prove hard math problems (Navier-Stokes existence+uniqueness, P/=NP, etc....)
Reading this
https://t.co/tOvR1LbfZ6
about the use of AI by some DeepMind folks to help prove the Navier-Stokes existence and uniqueness theorem, motivates me to point out that a couple months ago I attempted to prove this theorem with help on details from a couple LLMs,
https://t.co/uPCZ9GRnz5
The idea for the proof was mine -- basically to use the mapping from HJB equation (which I have used a lot in recent AI work) to Navier-Stokes equation, so as to map existence/uniqueness results from HJB over to NS. But the details get quite involved and I was not finding time to work through them myself and LLMs were quite helpful.
That said, these are thorny matters and I can't say I'm 100% sure there is no gap... I had set this aside because working on AGI seemed higher priority. I thought about trying to make this into an end-to-end Lean solution but didn't have time to deal with it yet....
Around the same time I also leveraged some LLM help to fill in details for an idea I had about how to prove P/=NP using a variation on quantale weakness (a conceptual/math tool I developed as part of my AGI work, generalizing Michael Timothy Bennett's notion of weakness as an alternative to traditional Occam's Razor)
https://t.co/vKP3grPtbV
Again working toward AGI is time-consuming and I haven't had time to do a full formalization (though obviously LLMs help with that too..) and as with any monster proof like this there could be a gap I'm not seeing....
I have not been making noise about these efforts because I wanted to wait to fully formalize and check things, but now that there is noise from others about similar endeavors I figure ok whatever....
Part of my goal in playing w/ these tough math problems [my PhD was math though I rapidly diverged to AI] has been to think about how to effectively use our Hyperon proto-AGI system as a "controller" for LLMs in doing this sort of work. I.e. in this sort of work there is a role for a broad-thinking creative mind to come up with proof ideas, for an LLM to work details, and then for Lean or some other ITP framework to do fully formal checking. The role I am hoping Hyperon can play here is the "broad-thinking creative mind" .... But in these attempted proofs I've just posted, it's been me not Hyperon trying to play this role ;) ..](/_next/image?url=https%3A%2F%2Fpbs.twimg.com%2Fprofile_images%2F1104599021660696576%2F4D5iBfAi_400x400.png&w=3840&q=75 "About using LLMs to help prove hard math problems (Navier-Stokes existence+uniqueness, P/=NP, etc....)
Reading this
https://t.co/tOvR1LbfZ6
about the use of AI by some DeepMind folks to help prove the Navier-Stokes existence and uniqueness theorem, motivates me to point out that a couple months ago I attempted to prove this theorem with help on details from a couple LLMs,
https://t.co/uPCZ9GRnz5
The idea for the proof was mine -- basically to use the mapping from HJB equation (which I have used a lot in recent AI work) to Navier-Stokes equation, so as to map existence/uniqueness results from HJB over to NS. But the details get quite involved and I was not finding time to work through them myself and LLMs were quite helpful.
That said, these are thorny matters and I can't say I'm 100% sure there is no gap... I had set this aside because working on AGI seemed higher priority. I thought about trying to make this into an end-to-end Lean solution but didn't have time to deal with it yet....
Around the same time I also leveraged some LLM help to fill in details for an idea I had about how to prove P/=NP using a variation on quantale weakness (a conceptual/math tool I developed as part of my AGI work, generalizing Michael Timothy Bennett's notion of weakness as an alternative to traditional Occam's Razor)
https://t.co/vKP3grPtbV
Again working toward AGI is time-consuming and I haven't had time to do a full formalization (though obviously LLMs help with that too..) and as with any monster proof like this there could be a gap I'm not seeing....
I have not been making noise about these efforts because I wanted to wait to fully formalize and check things, but now that there is noise from others about similar endeavors I figure ok whatever....
Part of my goal in playing w/ these tough math problems [my PhD was math though I rapidly diverged to AI] has been to think about how to effectively use our Hyperon proto-AGI system as a \"controller\" for LLMs in doing this sort of work. I.e. in this sort of work there is a role for a broad-thinking creative mind to come up with proof ideas, for an LLM to work details, and then for Lean or some other ITP framework to do fully formal checking. The role I am hoping Hyperon can play here is the \"broad-thinking creative mind\" .... But in these attempted proofs I've just posted, it's been me not Hyperon trying to play this role ;) ..")
About using LLMs to help prove hard math problems (Navier-Stokes existence+uniqueness, P/=NP, etc....) Reading this https://t.co/tOvR1LbfZ6 about the use of AI by some DeepMind folks to help prove the Navier-Stokes existence and uniqueness theorem, motivates me to point out that a couple months ago I attempted to prove this theorem with help on details from a couple LLMs, https://t.co/uPCZ9GRnz5 The idea for the proof was mine -- basically to use the mapping from HJB equation (which I have used a lot in recent AI work) to Navier-Stokes equation, so as to map existence/uniqueness results from HJB over to NS. But the details get quite involved and I was not finding time to work through them myself and LLMs were quite helpful. That said, these are thorny matters and I can't say I'm 100% sure there is no gap... I had set this aside because working on AGI seemed higher priority. I thought about trying to make this into an end-to-end Lean solution but didn't have time to deal with it yet.... Around the same time I also leveraged some LLM help to fill in details for an idea I had about how to prove P/=NP using a variation on quantale weakness (a conceptual/math tool I developed as part of my AGI work, generalizing Michael Timothy Bennett's notion of weakness as an alternative to traditional Occam's Razor) https://t.co/vKP3grPtbV Again working toward AGI is time-consuming and I haven't had time to do a full formalization (though obviously LLMs help with that too..) and as with any monster proof like this there could be a gap I'm not seeing.... I have not been making noise about these efforts because I wanted to wait to fully formalize and check things, but now that there is noise from others about similar endeavors I figure ok whatever.... Part of my goal in playing w/ these tough math problems [my PhD was math though I rapidly diverged to AI] has been to think about how to effectively use our Hyperon proto-AGI system as a "controller" for LLMs in doing this sort of work. I.e. in this sort of work there is a role for a broad-thinking creative mind to come up with proof ideas, for an LLM to work details, and then for Lean or some other ITP framework to do fully formal checking. The role I am hoping Hyperon can play here is the "broad-thinking creative mind" .... But in these attempted proofs I've just posted, it's been me not Hyperon trying to play this role ;) ..
Building Beneficial AGI - CEO @asi_alliance @singularitynet, @true_agi , Interim CEO @Singularity_Fi, @SophiaVerse_AI, Chair @opencog @HumanityPlus @iCog_Labs
微软也在 Windows 11 中测试原生 NVMe SSD，启用后 SSD 随机读写性能飙升，部分硬盘甚至随机写入性能提升 85%。原生 NVMe SSD 指的是不经过传统 SCSI 转译指令，现阶段 NVMe 读写都需要使用 SCSI 转译因此性能受削弱，所以微软开发了原生 NVMe 驱动提升 IOPS 性能。开启方法：https://t.co/xQ0FZ2WY4J
独立科技网站 - 蓝点网 / 感谢关注 订阅频道：https://t.co/xzeoUEoPcU 联系方式：https://t.co/LJK1g3biPp
My https://t.co/TKaKxRJ3V5 site seems to be working pretty well now, with a huge amount of new content. Not bad for a few days on the side as I worked on other stuff. Around 33k lines of high-quality shell scripts and nearly another 30k of typescript/react for the webapp.
I really tried to make this as comprehensible to someone without computer knowledge as possible; all jargon has super detailed definitions included. And I did multiple rounds of "audits" using agent trying to simulate the perspective of someone who doesn't know much about technology, which you can see documented here: https://t.co/8EMqD2Uv5k https://t.co/PodsF1DwYA It's still not as easy and simple as I'd like, but it's miles ahead of other guides, which don't even try to target the people I'm talking about here. I'd describe these people as being "hungry but clueless." They want to learn so they can acquire these magical new capabilities, but don't know the first thing about how to do it other than using a pure slop factory site like Lovable. Now they have a chance to get going in a real way using real tools that can scale with them and be used to create serious software. And with enough grit and determination, you can get Claude and Codex to solve most problems for you.
YouTube搞的零代码小游戏工具：Playables Builder，也就是说除了视频，用户还可以“上传”游戏到YouTube了 Playables Builder基于Gemini 3，一句话、一张图或扔一段视频，即可生成一个可玩的轻量小游戏 生成后可直接挂在YouTube，观众无需下载点进去就能玩 这或许会催生出一个“游戏短视频”新赛道，出现一批专做小游戏的MCN #YouTubePlayablesBuilder #AI小游戏
目前开放了Playables Builder测试计划，创作者需要申请加入，貌似有区域限制 地址：https://t.co/MgznRV60RE
Notably, YouTube was one of the platforms explicitly targeted by Australia's recent social media ban for people under 16.
Co-founder of @MechanizeWork Married to @natalia__coelho email: matthew at mechanize dot work
和 Next.js 官方文档样式一样的 Next.js 中文文档站开源了。 开源地址：https://t.co/ecU77hwu4z 去年，我的朋友 @awei_dev 照着 Next.js 官网，复刻了一个一模一样的文档站仓库。 后来我用这套代码上线了 Next.js 中文文档站，网站还挺受欢迎。 今天 @awei_dev 把文档站代码开源了，包含翻译后的中文文档，大家点点 star，也欢迎 fork 去搭建自己的文档 👉 https://t.co/ecU77hwu4z
⛵️全栈工程师，出海开发者，Next.js 手艺人 ｜ Next.js SaaS 模板：https://t.co/YzCDmGttme ｜ Nextjs 中文文档：https://t.co/Ek37EFh3tI ｜ 开源：https://t.co/vMeadXy9OD