softwaredoug

Someone suggested this follow up question to ChatGPT (which still exhbits this problem). The answer is even more non-sensical > Will they have a 50m hose to reach my car once I get there? Almost ce…

I just got the “you should walk” result on ChatGPT 5.2

Frankly, I hope he maximized the amount of money he made. It's a once in a lifetime opportunity. And nobody knows where AI is headed or if OpenAI even will be in existence in a few years given th…

Indeed, coding agents took off because of a lot of ongoing trial and error on how to build the harness as much as model quality.

I literally had begun to wonder if OpenClaw had more of a future as a company than OpenAI

Fashion production is responsible for 8-10% of all carbon emissions https://www.ifc.org/en/insights-reports/2023/strengthening-s... …

Anything that relies on gov't can be undone by gov't. Or weaponized by gov't. We need resilience that's hard to regulate or undo.

This type of centralization breeds authoritarianism. See also the Iran protests. There’s too many single points of failure in technology. These systems become sources of oppression inevitably. How do …

Nobody knows Some hypotheses (A) AI helps most setup simple tools - even non technical people. But once you get beyond greenfield the cognitive debt builds so you can’t reason about it it. It’s possib…

Codex/Claude gather telemetry by default. That’s why they are subsidized. You’re giving them training data. If you start with everything on GitHub, with maybe some manual annotated prompts for fi…

Isn't the study a year old by now? Things have evolved very quickly in the last few months.

Even within AI coding how people use this varies wildly from one people trying to one shot apps to people being barely above tab completers. When people talk about this stuff they usually mean very di…

This mode of operating seems fairly alien to me. It’s like a big waterfall design. It’s rare you have all the requirements of an app known up front. It’s pretty rare that they’re known so well you cou…

Do you understand the difference between an administrative warrant and a judicial warrant?

One difference, this was in response to an actual search warrant granted by a judge. What DHS is doing are administrative warrants, with no judicial overview (unless you sue to stop them).

Basically they are issuing (administrative) subpoenas. When they go to court (at the expense of the account holder) they back down so they don't get ruled against / told to stop issuing thes…

Underrated is how much improving harnesses, not just models, has a lot to do with productive uses of LLMs at tasks like coding in the last year.

What's interesting is they convinced the agent to apologize. A human would have doubled down. But LLMs are sycophantic and have context rot, so it understandably chose to prioritize the recent in…

When I write book chapters I write, throw away, write, throw away. Mostly with no a-priori outline But eventually I get to a point where all the failed attempts crystallize and it flows out of me star…

The tenor of the conversation is different than you expected?

Someone should make a musician training tool called tenorsflow

Musk burns through people at his companies. This isn’t that surprising. They’ll trick someone else into working there.

The problem will always be the training data. We can have LLMs because we have the web. Can we get to another level without a corresponding massive training set that demonstrates those abilities?

A local model is a smaller open model, so I’d expect it to be 9 months behind a small (ie nano) closed model as a base assumption