toobulkeh

I used the new system tonight and it felt like a definite downgrade. Generated a few non-working basic apps, couldn’t handle CSS in a NextJS environment. Terminal context didn’t work. And it went back to not reasoning through the problem until resolution. And kept slowing down.

I’m assuming major release vs stable, but this is pretty lackluster so far. Switched back to Sonnet reasoning. Here’s to improving!

I’d assume RL would be baked in to the request structure. I’m surprised OAI spec doesn’t include it, but I suppose you could hijack a conversation flow to do so

What a great overview!

I’d love your thoughts on my addition, autolearn.dev — voyager behind MCP.

The proxy format is exactly what I needed!

Thanks

Definitely is a core feature of BAML. My main complaint with BAML is that it’s all or nothing. It’s very opinionated and we can’t get the benefits without the DX and vice versa. Separating this feature without requiring a DSL of model definition is a great add.

Maybe it’s not for common use? I could see this betting important in the intelligence community, for example.

High speed improvement (4x) with low quality loss (2%). Sounds promising.

This is just pedantic noise. "Don't improve the model by using code gen" - problem "Write better architecture" - solution

It's literally the same thing. Sure, OpenAI's branding of ChatGPT as a product with GPT-5 is confusing, because GPT-5 is both a MODEL and a PRODUCT (collection of models, including GPT-5).

But does it matter?

I don’t know (yet) but some ideas might be: 1. A skill that searches skills 2. MCP that returns only relevant skills based on context 3. Separate concerns through sub agents that each have their own autolearn server

Git repo: https://github.com/tarkaai/autolearn

It’s like 5 hours of work and I’m super early with it!

Definitely needs some work here. That isn’t done (yet). Great feature request! I think it would be some kind of testing logic or feedback loop. Or both!

Demo video: https://youtu.be/s_9m4P9_6jc?si=P7By-zl62GkFMRpH

Any benchmarks or comparisons on improvements?

I vibe coded this at a hackathon a few weeks ago. The solution isn’t stable yet, but it’s similar to Claude Skills with the benefit of only needing an MCP client.

I open sourced it too.

It has a bit to go—like automatically monitoring reasoning traces and security—but I thought this would be easier to use than trying to maintain your own skills.md collection if the server handled more of it for you.

I’d be curious if anyone thinks I should keep working on it. I’d love someone to collaborate with on it!

I implemented a rudimentary version of this based on some BabyAGI loops, called autolearn: autolearn.dev

I love this per-agent approach and the roll calling. I don’t know why they used a file system instead of MCP though. MCP already covered this and could use the same techniques to improve.

Jumping up and down in front of another messenger and getting a human reaction made my jaw literally drop.

It was the cherry on top of this complete flex of brain power. This is like Pixar’s demo reel. Could easily define a generation of technological advancement.

#StopKillingGames

Did you miss the entire article? CS itself came from a dorm room. You can have excellent games that spawn from creativity instead of monetization.

I was really into the odd maps (NIPPER) and early Internet community around games (joe2). We hosted servers off of unused CPU cycles from oil exploration boxes.

This still exists all over the web, but the creators that figured out economics moved on.

Why ditch Chromium browsers? Forks can still get away from the Chrome specialities. (Eg Brave)

It has nothing to do as a “gotcha”. It’s about improving error codes and other interactions for agentic editors.

Yes, the books are much faster and enjoyable than the movies. They reveal a lot more of the WHY and it’ll help you see what has already happened in a new light.

The TV series has included extra things that the books do not have, which haven’t added much to the plot, just a slower pace.

I’ve been tinkering with this, but haven’t found a pattern or library of someone solving this.

Have you?

I think financial damages are realistic here. They materially damaged another brand, it’s all business at the end of the day.

Enshittification. The UX is not for you—it’s to maximize you.

We're on sentry today, but have been waiting for a fully OSS solution like this.

Or you can ask a GPT, that has already indexed your publicly available support docs, to prioritize potential places for a user looking to keep access as a backup.

AI is available to everyone, and we’re not prepared.

I’d love to see better optimized web socket support and “save” features that cache LLM queries to optimize fallback

Not just any high school. https://exeter.edu/

I agree with most of the sentiment with his post, and was enlightened to learn about his perspective on universities and research into the history of it, but the argument of language nuance itself (colored vs PoC) feels short sighted.

All language has nuance. And the language is very high on the Maslov’s hierarchy, but that’s the point. It’s a progressive discussion. Terminology has meaning and we’re growing our understanding what the meaning is.

You can have discussions and understanding or no understanding and ignorance. The problem is not the language or understanding, it’s the actions itself. Being aware and understanding is not bad. The answer is not the counter culture naivety or cancel culture.

Yes, cancel culture and prigism are abominations of high society. That’s the action. In the same turn, the term “woke” has been absolutely weaponized from the counter culture point of view. We, as a society, are figuring out its place. It’s definitely not in public schools or politics. It has, like abortion, been co-opted for power in a democratic society. Let’s focus on the action and changing our systems to enshrine cultural norms (like a public service of unbiased news) into law instead of relying on the markets.

There's definitely not enough of these solutions being open-sourced. Definitely makes it more accessible to compliance or restricted environments. More AI should go this route. Definitely will be trying it out in my next build.

I don't disagree with his statement, but I disagree with his expectations. These systems are NOT ripe for capitalism. Capitalism is built to fulfill demand. There was never demand from the government for it to distribute cash, therefore they never built a system for it. There was never a real demand for a vaccine, so we never developed it.

Capitalism is why we're in this mess - we're valuing short-term gains over long-term/far-fetched/society-changing results - and a16z is part of that system. Sure, he's at the edge of that system, but I guarantee you he would not have invested in a coronavirus vaccine in mid-2019 when there was no market for one.

I would not say it's time to build. I would say it's time to change our system of motivation. Money does not, and will not ever, solve these problems BEFORE they become problems.

Zendesk has a decent Knowledge base system. Intercom is slick, but expensive if you're trying to be frugal. Discourse is the best forum on the OSS market.