Igor Adamski

building

[Igor Adamski @igoradamsk]

@WJongert Ur speaking to the fuckin void dude

[Igor Adamski @igoradamsk]

@cursor_ai can you go back and allow me to resize the chat and widen it?? s😭😭bbing

[Igor Adamski @igoradamsk]

@HMRCcustomers I'm meant to receive a cheque from you because I paid too much tax. I moved to another address and only changed it yesterday (got the email from you on monday) - could you check and make sure you send the cheque to the new address?

[Igor Adamski @igoradamsk]

@nsbradford @datadoghq @cursor_ai I added the datadog MCP to the https://t.co/T5Bz2RAPjV with the DD_API_KEY and DD_APP_KEY in the headers (server URL: https://t.co/6KxGpj9hQE) and the agent says it doesnt see this MCP. Anything I'm doing wrong? Thanks!

[Igor Adamski @igoradamsk]

@nsbradford @datadoghq @cursor_ai Doesnt work yet I’m afraid

[Igor Adamski @igoradamsk]

@nsbradford @datadoghq @cursor_ai I just want to live in a world where I "@" cursor in Slack to investigate an issue and it uses the Datadog MCP for it. I think the intention is clear and as far as I understand you guys are working on that already :)

[Igor Adamski @igoradamsk]

@nsbradford @datadoghq @cursor_ai I can login, but I wanted to use the Automations to hook it up whenever our users report a problem

[Igor Adamski @igoradamsk]

@nsbradford @datadoghq @cursor_ai I wasnt able to set it up for the Cloud Agent as it needs me to open a webpage and log in - is there a way to bypass that?

[Francisco Moretti @franmoretti_]

So glad to work at a place with no token limit

If you care about this, reach out. We are hiring in Europe

[Francisco Moretti @franmoretti_]

Stop rebuilding the same AI chat infrastructure.

Introducing ChatJS

[Igor Adamski @igoradamsk]

thinking of making this change what yall think

[Andrej Karpathy @karpathy]

@burkov The code was written at layers 22-30 and is stored in the value activations you just can’t read it. I think you owe the LLM an apology.

[Igor Adamski @igoradamsk]

@agupta first, you need to get the "trillions of tokens" and then you need to have relevant data for fine-tuning. also, you never train just one model - its a experimentation process. all possible - but I doubt the total cost is <$500k

[gabriel @gabriel1]

sf has the most autistic billboards

[Andrej Karpathy @karpathy]

In era of pretraining, what mattered was internet text. You'd primarily want a large, diverse, high quality collection of internet documents to learn from.

In era of supervised finetuning, it was conversations. Contract workers are hired to create answers for questions, a bit like what you'd see on Stack Overflow / Quora, or etc., but geared towards LLM use cases.

Neither of the two above are going away (imo), but in this era of reinforcement learning, it is now environments. Unlike the above, they give the LLM an opportunity to actually interact - take actions, see outcomes, etc. This means you can hope to do a lot better than statistical expert imitation. And they can be used both for model training and evaluation. But just like before, the core problem now is needing a large, diverse, high quality set of environments, as exercises for the LLM to practice against.

In some ways, I'm reminded of OpenAI's very first project (gym), which was exactly a framework hoping to build a large collection of environments in the same schema, but this was way before LLMs. So the environments were simple academic control tasks of the time, like cartpole, ATARI, etc. The @PrimeIntellect environments hub (and the `verifiers` repo on GitHub) builds the modernized version specifically targeting LLMs, and it's a great effort/idea. I pitched that someone build something like it earlier this year:
https://t.co/ANHhasxzD8
Environments have the property that once the skeleton of the framework is in place, in principle the community / industry can parallelize across many different domains, which is exciting.

Final thought - personally and long-term, I am bullish on environments and agentic interactions but I am bearish on reinforcement learning specifically. I think that reward functions are super sus, and I think humans don't use RL to learn (maybe they do for some motor tasks etc, but not intellectual problem solving tasks). Humans use different learning paradigms that are significantly more powerful and sample efficient and that haven't been properly invented and scaled yet, though early sketches and ideas exist (as just one example, the idea of "system prompt learning", moving the update to tokens/contexts not weights and optionally distilling to weights as a separate process a bit like sleep does).

[Igor Adamski @igoradamsk]

@romdot2 The next step towards singularity is only reading an executive summary of the changes + steps taken to ensure everything works

[Igor Adamski @igoradamsk]

I feel like a boomer figuring out why I replied to myself instead of continuing a thread

[Igor Adamski @igoradamsk]

At the end of the day, it’s just HTTP calls + managing state/context.
You should control that however you feel best.

[Igor Adamski @igoradamsk]

Opinion: LLM frameworks (LangChain, LangGraph, etc.) are dead weight.
🧵

[Igor Adamski @igoradamsk]

The overhead of building your own orchestration? Tiny.
The cost of depending on a slow, bloated framework? Massive.