teaearlgraycold

Don't talk to the police!

The US constitution should require proof of harm to a living thing as a prerequisite for criminal charges.

What’s up with the cop that cited them for that?

I really need these LLMs to start being more like the gate keeping assholes of StackOverflow. Just a little dash of that energy would go a long way.

I asked about Tiananmen Square and it said it knew nothing about it. A real life “Doesn’t look like anything to me” moment.

They're great at helping me look up web dev stuff when I don't have internet access.

In my experience GLM 5.2 is a pretty good Opus replacement. But K3 has not given me an experience on par with Sol or Fable. The price/intelligence ratio might still make sense. But it’s not very inspiring when it comes to my real world tasks. I’m doing pretty mundane web stuff.

If we’re talking about human technology available in a few hundred years, don’t discount far more exotic options. I’ve heard people talk of theoretical terrestrial lasers pushing on tiny probes. With an absolutely gigantic laser and magical material at the back of the probe that won’t instantly vaporize there’s enough energy to get something the size of a smartphone up to a reasonable proportion of the speed of light.

I can’t prescribe this theoretical technology to the problem. But I also think it’s unreasonable to set the limit using known technology and then discount the idea altogether. We have no idea what will be possible in 300 years.

For example you cannot have concurrent access. As soon as you need a worker process and a web process SQLite is out. Or if you are trying to use it as a vector db all of those vector searches will block a node event loop.

An acoustic analyzer for a sink basin in a hackerspace. It will allow for automatic detection of dishes/cups left in the sink without cameras and can be handled by an rp2040. So far looks like it has very low false positives but can only successfully detect certain materials or large masses.

It works really really well. No one should be writing untyped JS anymore.

Sad to see no mention of @ts-check. I love being able to write native browser JS modules with nearly all of the functionality from TypeScript.

Well yes of course the non compete is a problem.

Cannot pursue your vocation in certain jurisdictions. And there are plenty of adjacent fields I suspect wouldn't be covered by the NC - like education.

Clearly something else was going on. No mentally healthy person would kill themself because of this.

At least in CA they need to disclose them now. Previously some restaurants would just hit you with a surprise mandatory tip.

I read at talking speed so not as much to gain with text.

"just here for the paycheck" vs. "here for the paycheck but also genuinely enjoy parts of the job" are miles apart.

What's an example of what the 10 agents would be doing in parallel? I've never used more than 2 in parallel before.

Charging by line of code is crazy

I feel like the world cares more about stopping the spread of CSAM than it does the actual abusive actions against children.

Trying to imagine a world where I use Chrome unironically.

RAG, as far as I understand, is a term that came about before LLM tool-calling was as prevalent. Your options were to have an LLM hallucinate up a response, or instead do a [document -> chunk -> embedding -> vector db -> query -> context window] pipeline. I haven't heard anyone talk of LLMs + web search or other tool calls as RAG, even though if you pull apart the semantics the term is applicable. In fact I don't hear people talk about RAG much at all. I suppose much of what people were trying to solve with document chunking/embedding pipelines has been solved with bigger models and tool calls. And along with that change in tooling we have left behind the term "RAG", which leaves it attached to the concept of those pipelines.

Just don't use a Twitter post as your demo video in the GitHub readme. I don't care what someone does on Twitter - I never go there. But you can just embed a .gif or .webm in your readme or link to YouTube there.

I kindly request people do not link to Twitter. I don’t have an account and won’t make one to see a video clip.

It’s tough to write good questions for LLM evaluations. They’re so good at picking up subtleties they can pass a multiple choice test when given only the answers and not the questions.

People optimizing for grades is so sad. I got a BS in CS and no one has ever asked about my grades. I tried to optimize for learning and it turns out that pays off really well once you hit the real world.

I suspect LLMs would do a good first pass with function naming.

AI can be anything from depth first search to superhuman intelligence on a chip. I like to refer to it as automated statistics.

Here’s my attempt:

Assume you have a fan sitting still. You smack it and it’s now rotating with 1m/s angular velocity. If you want it to go faster you can’t smack it at the same speed. You have to hit it faster else you’re just tickling it and it stays the same speed. So you smack your hand twice as hard and now it’s going even faster. Then three times as hard, four times, etc.

If you sum the smack energy it will be 1+2+3+4, which starts to build out a right isosceles triangle if you graph it. Such a triangle is half of a square, ie: 1/2*v^2.