10001110101

Used Teams for a bit. Seemed fine, just used it like any other IRC clone. Didn't use it for video. Windows has a lot of annoyances; death by a thousand cuts. The Windows ecosystem also sucks: to the point where graphic card and mouse driver installers try to install spyware.

I think there's a massive oversupply of software engineers world-wide, and investors and executives are heavily pushing offshoring to countries where there are even more engineers that are even more desperate to find work. The ideology or focus of the entire US investor/executive class seems to have shifted as soon as Musk gutted Twitter. I fear this may be another, "these jobs aren't coming back," kind of thing the manufacturing industry went through. Perhaps we'll see a boom of bootstrapped start-ups ran by engineers (or preferably worker-cooperatives), but that's extremely hard to do.

I think tool calling is just a form of structured output, where the LLM outputs something like json that describes the function to call and the arguments, then you parse that and run the function with the arguments, then feed the output back to the LLM in a new message, if you want. IDK the specific details, I'm guessing there are some special tokens some LLMs produce for tool calling, and I'm also guessing there is "controlled generation" (masking the logits/tokens, and only choosing to generate the tokens that would be valid). Ollama apparently doesn't support the special tool-calling tokens or output structure that some models use.

This is how it looks when using OpenAI compatible APIs: https://platform.openai.com/docs/guides/function-calling?api-mode=responses.

I've never heard of TabbyAPI (I'm new to using local models, in general). I'm also not sure what TabbyAPI brings over Ollama, llama.cpp, or vLLM. So would be curious as well.

Edit: For my little toy project I'm working on, I'm switching to not using tool-calling at all, and building a LangGraph and using structured output, which should be more reliable with my use-case. I.e. just always call the tool manually, feed the output back to the LLM to have the LLM evaluate if the output was correct, retry calling the tool with different arguments if not, and just fail after 5 calls.

I like watching Yannic Kilcher, to keep up with some of the newer developments, and read various papers. I do have a background in ML (mostly the more traditional, non-generative, supervised learning and reinforcement learning) though.

He absolutely believes in those things, and has shown multiple times he's willing to light his money on fire to support his ideals. He's idealistic (adjacent to the neo-fascist Dark Enlightenment movement), and not just simply motivated by wealth.

I think this is only a small part. Interest rates are kinda high. VCs only want to invest in companies with AI exposure because of all the hype. From companies I've interviewed with, offshoring seems to have accelerated dramatically (companies only had or wanted a few US devs to manage larger Indian teams). I've visited career pages of companies working in the business domain I have the most experience with, and all open software positions are exclusively in India.

I have over a decade of experience as well. Nobody in my small personal "network" knows anyone that's hiring right now (I hate the fakeness of networking for networking sake, and am not very social, so I don't have much of a network). I've applied to hundreds of job postings over 6 months, interviewed with maybe 6 companies, and rejected usually just because they were also interviewing 10-20 people for the same role, and another person had slightly more experience with a specific part of their stack, or they just liked another person more for whatever reason. I believe all remote job postings get 1000s of applicants, and every one local to me get 100s.

It all kind of reminds me of when I tried using online dating apps, lol.

https://deusex.fandom.com/wiki/FEMA

They work for the shareholders, not the customers. For most publicly traded companies, the stock is completely detached from fundamentals, so they just do whatever the large investors like (often just hype the new hottest thing; such as marketplaces or "increasing efficiency" with layoffs), regardless if its good for the "real" business or not.

Idk, the author of "Civil Disobedience" did it. It's a valid resistance method, or just a way to keep yourself morally consistent. Every kind of resistance has a good chance of bringing harm to yourself.

I personally didn't file one year; kinda just kept putting it off because I knew I wouldn't be able to pay my taxes (was a 1099 worker, and suffered some major financial blows that year); nothing ever happened, but it definitely was a risk. I probably won't do that as an "act of resistance" next year, but will probably try extending as long as possible.

Hmm, Devstral doesn't call any tools for me in the current stable Ollama version or the current release candidate. Wonder if it's a bug in ollama or langchain. I've since tried "QwQ-32B-GGUF:Q3_K_XL", and it's a little better than Qwen3-14B:Q6, but still not quite satisfactory, and is much slower and "thinks" too much.

We let the rich, who aren't necessarily the smartest people, get too rich, so they gained too much power. Worldwide problem though, as their influence doesn't stop at borders.

Kinda weird GPT4-Chan wasn't referenced. A guy fine-tuned GPT-J on 4chan, then deployed bots to write posts. I guess it was more of a stunt than academic or scientific, but training on 4chan improved the model's performance on a truthfulness benchmark.