SwellJoe

25 ms·

601.
by SwellJoe 5mo ago
What's political is a policy change to "fast track" companies into the Nasdaq 100. Spacex is the first to benefit from this loophole that allows them to be added to indexes almost immediately after listing, which likely is a
602.
by SwellJoe 5mo ago
You shouldn't have to sue to have contracts honored by large corporations, and it's entirely reasonable to shame the fuck out of a corporation that responds to a situation like this with "fuck you, sue us if you don't li
603.
by SwellJoe 5mo ago
This is comic book villain shit. The story isn't told well anywhere that is covering it, YouTubers aren't always exactly great journalists or incentivized to tell a coherent and concise story. But, it's obvious the Bricks and
604.
by SwellJoe 5mo ago
And, they're starting to push AI music into recommendations, as well. Their incentives are not aligned with users or artists. The major labels were previously pretty OK with the situation, as they had a negotiating advantage against in
605.
by SwellJoe 5mo ago
I can't imagine the mind that wants more facebook.
606.
by SwellJoe 5mo ago
You need a huge corpus of training data for the language for models to be good at using it. Rust has that (so does Go). OxCaml probably does not (unless there's some iceberg of open code out there that I'm unaware of). I'll t
607.
by SwellJoe 5mo ago
Age check is identity theft at scale, mandated by the state. A disaster waiting to happen (and it won't wait long).
608.
by SwellJoe 5mo ago
"Cold outbound"? So, spam?
609.
by SwellJoe 5mo ago
Nothing says freedom like being forced to go to a re-education camp.
610.
by SwellJoe 5mo ago
Whenever I've seen anyone suggesting a wealth tax, it is specifically to address the very wealthy who pay an effective 0% tax rate, because they use the "buy, borrow, die" strategy. These are not wage earners, working a regul
611.
by SwellJoe 5mo ago
It's not clear to me what this does that tmux doesn't? Agents can already interact with tmux, so what does this provide?
612.
by SwellJoe 5mo ago
That was a surprising sentence. But, Antigravity is fine. I mean, I only open it because it's the only way Gemini is going to get tool use right on the first try, but it works like all the other VS Code forks (acceptable, not great). I
613.
by SwellJoe 5mo ago
I've turned off memory features in everything because of this kind of problem. It makes models dumb as hell. And, it makes them more effective sycophancy machines, which are probably unhealthy to interact with. I doubt the OP is actual
614.
by SwellJoe 5mo ago
You're missing an agent. The model uses tool calls to interact with the filesystem, commands on the system, optionally search (you need a search MCP server, like Brave or Exa, and API key), etc. I usually use the Zed Agent built into Z
615.
by SwellJoe 5mo ago
No, I've just seen benchmarks showing most models start degrading around 4-5 bits. That's not to say they become useless, just that down to about 6-bits (with careful hybrid quantizations like unsloth where some of the layers aren
616.
by SwellJoe 5mo ago
Two 3090s is 48GB, so it's possible to run the 6-bit quantization comfortably, which is fine. It doesn't start to get notably dumber until lower than that. It won't be as fast as a hosted model, but dual 3090s will be comfort
617.
by SwellJoe 5mo ago
I would disagree. I think it'd take a lot of prompting to make GPT 5.5 not have the underlying personality of GPT, which I find awful. They have knobs in ChatGPT to choose a "professional" tone, which improves it somewhat, bu
618.
by SwellJoe 5mo ago
That's a different problem than I thought you were worried about. I wasn't considering the marketing angle, though that is certainly relevant and a risk to consider, especially when it comes to Google, whose primary businesses are
619.
by SwellJoe 5mo ago
We're having DeepSeek moments every couple of weeks. Qwen 3.6 hit hard in the self-hosting space. It's incredibly capable for its size, really shaking up what's possible in 64GB or even 32GB of VRAM. The Prism Bonsai ternary
620.
by SwellJoe 5mo ago
I have and use both Claude Code and Gemini CLI, and still don't consider Gemini worth starting for coding except to review Claude's output in critical commits (on a security boundary, maybe broad refactors, etc.), though I try sid
621.
by SwellJoe 5mo ago
At least in some cases, there seems to be a move toward training on more synthetic data and strictly curated data, especially for smaller models where knowledge can't be extremely broad, because there just isn't enough room to sto
622.
by SwellJoe 5mo ago
Opus 4.7 is smarter than even Gemini 3.1 Pro on nearly every metric, though. You're comparing apples to oranges. Gemini 3.1 Flash is somewhere in the neighborhood between current Haiku and Sonnet, I think? Still a better value than the
623.
by SwellJoe 5mo ago
That's a lot. DeepSeek v4 Flash is just over a tenth the price, and DeepSeek v4 Pro is roughly the same price (currently heavily discounted, but will be $1.74). I mean, the benchmarks for Gemini 3.5 Flash are very strong, but at those
624.
by SwellJoe 5mo ago
I don't know where a vengeful future administration would come from. We only have Democrats or Republicans to choose from, and Democrats have made turning the other cheek their entire purpose and political mission. They slow-rolled the
625.
by SwellJoe 5mo ago
They made it less sycophantic. Which is a good thing for mental health, but maybe a bad thing for popularity contests.
626.
by SwellJoe 5mo ago
With an Austrian accent, perhaps?
627.
by SwellJoe 5mo ago
I've come to be convinced that having a huge amount of money causes some kind of mental breakage, a need to control other people that is unhealthy for everyone it touches. I don't mind everyone having or expressing an opinion, eve
628.
by SwellJoe 5mo ago
Everything is currently heavily subsidized. If the AI companies don't improve efficiency, they'll eventually have to start charging what it actually costs to offer the service, which is a multiple of what the currently charge. I e
629.
by SwellJoe 5mo ago
That's just the plain text (or whatever files), that's not the context the model is directly working with on the server, which is tokenized, embedded, vectorized and has attention run against those vectors. The local history is ge
630.
by SwellJoe 5mo ago
The context window is not on your system. It's on the server with the model. There may be some local prompt caching, of some sort, but you're not locally hosting the context unless you're also locally hosting the model.

More ›