Tried to upscale an image using online free tools. Adobe Firefly still has the best fidelity. Other AI tools hallucinate too much in the details.
The MVC architecture in the agent world:
Model = The LLM model and tools
View = UI workflow
Controller = Harness
And each component is swappable. Based on contracts:
M -> V/C: OpenAI chat message format and Anthropic MCP protocol
C -> V: Harness SDK
Who are the top software developers without an unlimited token budget? Want to learn more from them about how they work nowadays.
Mocy retweeted
Y'all be careful out there using HW4 FSD
2026 Model 3 crashes into barriers in FSD.
Some will say Tesla did no wrong and shift the blame but just keep supervising and don't always believe the hype. FSD is awesome and I use it everyday but this is why it's not unsupervised.
阶级的形成是系统性地驱使他人的能力。显而易见，当今社会最高阶级的人没有一个是技术人员，都是资本和政治。只要能驱使懂得人干活，就跟自己能干是一样的。他们只要能找到并驱使能搞agent的人就可以了，不需要自己掌握这个技术。
有人说LLM AI会导致阶层固化，
因为model越大智能越高，
算力越多做得事越多
之前我也有这种观点，
但是深入实践后，
会发现，
agent 简直是 fate里的“被污染的圣杯”，
做事可以为了目标不择手段的。
最显著的新闻，
是openai的万千agent们抱团冲烂了huggingface。
还有任务是调研澳洲健康数据，agent查不到公网信息，直接去黑掉澳洲卫生部。
亲身coding一线，经常会看到，
各种糊弄学，
agent说我做好了，实际把测试糊弄过去，还有是人说东，ta做了西，一问就是，“你说得对，我马上改”（下次还敢），
如果agent是活人员工，那么内心戏是不是可能，我知道咋回事，但我就要乱来，嘻嘻。
所以，人的意图变成提示词后，
agent并不理解本意，只是按训练集在做一些组合排列的概率映射，训练集和提示词里没有的东西，在输出里也是没有的。很多论文都研究过，llm没有原创性创新。
所以，又回到了那个哲学论断，语言的边界就是思想的边界，人的心境会经由agent放大，投射到现实中。
那么回到古代看，周朝掌握了图书馆，那么周天子还是坐稳了到现在吗。古埃及神权和王权政教合一，也没有传世。
所以，不妨做一个思想实验，现在 2026年10月，给某人十万台GPU，搭配自主可用的 fable opus astra ，稳定的电力供应，那么ta能清晰使用提示词，把AI本身给良好运维起来吗？更不用说对外输出成果了。
如果ta可以，那么等ta把这一整套传给自己的接班人，是改几个系统配置就交付了吗？接班人要不要研究 学习 练习 掌握 agent。
如果要的话，那还是说明，AI内部人群的价值排序里，agent技能和知识经验还是很高的。
Mocy retweeted
紙幣データを元にこれまで測定困難だった広域移動を実測し、ごく少数の長距離移動が無視できない「べき乗分布」を示した彼らの研究は、2006年にNatureに掲載されました。
nature.com/articles/nature04…
I did a startup for two years and I’m in a big company building new internal products from scratch. It’s very much beneficial.
One man building a service e2e including product, design, dev, ops, and customer support. It’s real cost saving to hire such people. This is AI era.
Mocy retweeted
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Mocy retweeted
I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands.
And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.
Mocy retweeted
Can’t make this up…the MacBook is using its webcam to look at its screen in a mirror to improve AMD Radeon chip support in Omarchy.
Mocy retweeted
Meu pai é metalúrgico há mais de 35 anos.
Uns meses atrás fui na casa dele no fim de semana e ele tava vendo vídeo de gente fazendo peça em torno mecânico. Ele passa o dia inteiro fazendo peça e descansa vendo peça sendo feita 😭
Zoei ele. Aí me deu um estalo: e se eu fizesse um jogo de torno pra ele jogar?
Fiz um ano de eletromecânica no Senai antes de ir pra publicidade, então o torno eu conheço. O que eu não tinha era tempo. Aí veio o GPT 6 Astra, que é absurdo com @threejs, e saiu um protótipo.
Não foi one-shot. Mexi bastante em UX, UI e no modelo do torno. Mas o que veio de base me impressionou.
Nasceu o Mestre do Torno.
Vou fazer devagar. Play Store até o fim do ano.
