Discerner

@birch_js Ahhh! I take that back... This config file may not be git cloned or git pulled (i guess) ?

@birch_js Great! now I don't need to require nodejs in my java project just because I need decent pre-commit hooks

Hey, maybe we don’t need husky anymore!

Hermes Agent v0.11.0 - “The Interface Release” Full changelog below ↓

@Teknium @NousResearch This is amazing 🤩 Happy to see that one of the merged PRs is mine 🙂🙂🙂🙂

@NousResearch Our biggest release yet with over 700 PRs merged

@kiran27rathod @VishalMalvi_ Yes correct. things like DV Act, BNS 85 and much more fall into place place

@VishalMalvi_ Being a man...if a female can do this to you....it's a shame.....2 slaps...3 punches and things fall into place.....man need to be man enough

After being abused for the past two years bro recorded these clips on his laptop Think twice before getting married 🙏

@svembu Wow! The Roneo machine memory hit me very vividly.

In the 1980s, most IITians would go abroad. In 1989, when I graduated from IIT Madras, I remember feeling extremely dejected about our country. Punjab, Kashmir and Assam were all burning. My heart was not in engineering. I was mostly reading books in Economics and Philosophy - we had a good library. The burning question in my mind was "Why are we so poor?" Some of my classmates and I wrote an article in the IIT campus newspaper in late 1988-early 1989 (there were two newspapers, Focus and Spectator, and I believe we published in Focus, they were reproduced using "cyclostyling" machines - please look them up!). In my vague recollection, the thrust of the article was that the IIT system was failing to serve the needs of the country and the country itself was facing a profound stagnation (I wish I could get that article now - a copy may be in some dusty basement in IIT). I want to know what I thought and said as a 21 year old in 1989 that I agree with and what I disagree with today. By 1989, I had become a committed anti-socialist, having lived through the socialist stagnation of India. By 1989, the collapse of the Soviet Union was on, and China was in turmoil - the Tinananmen student protests and their forced suppression. By 1991, India needed an emergency IMF loan. The 1991 economic reforms by Shri Manmohan Singh happened due to pressure from the IMF. So you can imagine the mood in 1989. That was the India I left in 1989. I was feeling miserable to leave but hopeless to stay. In 1990, I came home for a visit and thought of dropping out of my PhD and staying home. I was home sick. I started to study Singapore and Japan during 1990-94 in my PhD years - the "Why are we so poor" question. By 1994, I decided I would be in the private sector and took up an R&D job in Qualcomm.

@aravind what is your definition of AGI? Each person has their own definition

I predict Grok to near AGI in late 2028 based on current trends and compounding acceleration. Grok 5 should be 8-10X Grok 4 in capability and I estimate it released by end of this year in 2026. By the end of 2027, we must have truly AI assisted AI research and development. This means the improvement in AI models will be high degree exponential. Compute is also scaling exponentially. So Grok 6 in 2027 may achieve at top 0.1% of human ability in research and innovation. And Grok 7 in 2028 will be capable of running its own R&D, invent new science, and self improve. So by end of 2028, very high chance that Grok will be near AGI. Now will be it be released in public or kept under wraps due to Govt pressure? I cannot say.

@XperienceIA 18th April - Did you get in too??

@Discerner4u Quando você fez aplicação? Estou aguardando aqui, mas ainda não recebi nenhum e-mail. Parabéns!

Yaay! My idea got selected for Claude Code Hackathon - Build with Opus 4.7 🤩 Wish me luck 🙏🏽

@akshaymarch7 I somehow feel it is the opposite - it enforces the illusion of competence. Such people are not afraid of it though. They love it.

The real reason many fear AI is because it removes the illusion of competence.

@Teknium @austinbv Yea it adds confusions around which model is used when. Also smaller prompt may require more work too.

@austinbv Its a rudimentary not recommended thing that people demanded. I never believed in model routing for tasks. I'll remove

How are people using smart model routing on Hermes? It seems to lobotomize itself when it's not using opus. @Teknium ??

@AbhinavXJ Do i see Mr. Dorsey there? 🤯

yc hired south indian movie villains to keep us silent bro 😭😭😭😭 wtf

This is why Opus 4.7 is 7.5x

Link to paper: zenodo.org/records/196401…

A new fresh paper is out - Agentic coding for Brownfield systems! That too, from a workflow that was introduced in 2002! Link to paper in the thread 👇

@burkeholland Shouldn't the model seller tell what it is good for? And how it is better than earlier version?

Impressions of Opus 4.7 so far are that it is a very expensive 4.6, which is an expensive 4.5. I don’t see the difference. But to be fair it takes some time for us to see where these new models really shine.

@Teknium Thanks for the clarification. Just curious - is there a specific constraint driving this? What part needs this specific number as the bare minimum? Sorry I am asking too many questions

@Discerner4u Hermes really will just not function appropriately if you cant get to 64k context. I'm sorry :/

@Teknium I seem to struggle with the minimum requirement of 64k context window by hermes. I know it can be changed in config. Does that change how my agent worked before if i use the same model? I am using a local model with a constrained GPU for context.

@burkeholland Welp! Inflation is real!

well hello little (pricey) fella

@unclebobmartin @wookash_podcast Wouldn't this affect us badly in the eventuality of funding drying up for these AI companies? Or, when we move to the era of AI being too costly for unfunded opensource devs? I am not ready to lose my code-reading and code-writing skills until I am sure they won't be needed.

I don’t review code written by agents. I measure things like test coverage, dependency structure, cyclomatic complexity, module sizes, mutation testing, etc. Much can be inferred about the quality of the code from those metrics. The code itself I leave to the AI. Humans are slow at code. To get productivity we humans need to disengage from code and manage from a higher level.

To people who are *good* at reviewing code (or claim to be hehe) - how is that possible? To what extend you can properly review the code with low familiarity with the codebase? Eg. New project, you jump in, Claude Code PR - 500 lines changed - review now What's the strategy?