SwellJoe

24 ms·

481.
by SwellJoe 4mo ago
I guess. But, I have a hard time believing that market is large enough, or wealthy enough, to support the custom manufacture of a luxury dumb phone. I've been wrong about products before. But, I've been right more often. I'm
482.
by SwellJoe 4mo ago
I have no idea who the customer is for this product. I bought a C64U. I bought a MEGA65. I bought the T-shirt. I own nearly every original Commodore 8-bit. If anybody is the target market for Commodore products, I am the target market, and
483.
by SwellJoe 4mo ago
> If you want a smarter AI, it needs to have not-dumb memory. Who says? According to what metric? How would you prove that assertion? > Now the memory isn't dumb, so the results are better. (And the article shows you why it'
484.
by SwellJoe 4mo ago
> but it's much smarter than "shove markdown files into directories". Is it, though? I mean, is there evidence "bunch of markdown files" is bad while "database the model has to be instructed how to use"
485.
by SwellJoe 4mo ago
I stand corrected. But, in my defense, I did say "software that could be useful to folks with accessibility needs that don't try to address it (within bounds of their resources and capabilities, obviously lots of OSS just doesn&#x
486.
by SwellJoe 4mo ago
Literally none of the examples of large companies using it in the comment I replied to are game companies or using it for game debug menus. You made up a guy and pretended I was yelling at them.
487.
by SwellJoe 4mo ago
OK, if "toy project" isn't the right word, then perhaps, "unethical" or "exclusionary" would be better words to use. I judge software harshly that could be useful to folks with accessibility needs that d
488.
by SwellJoe 4mo ago
Gemma 4 is competitive with Qwen 3.6. I had vague feelings that Qwen was better at coding tasks, based on anecdotes and public benchmarks, but I've been doing some benchmarking lately, and Gemma 4 31b is consistently beating Qwen 3.6 a
489.
by SwellJoe 4mo ago
I added it to my benchmark based on Mythos-reported bugs, and it's better than GLM 5.1, but still behind several other models, maybe most directly comparable to Qwen 3.7 Max. But, several other open models, including small self-hostabl
490.
by SwellJoe 4mo ago
I added this to my benchmark of models looking for Mythos-reported security bugs. Unsurprisingly, it found 0. There is, after all, a lower bound on how small a model can be and still find security bugs. https://swelljoe.com/
491.
by SwellJoe 4mo ago
Everyone I work with who used Cursor stopped using Cursor when Claude Code came along. They're back to their regular IDE when the need to read code, or they just review it at PR time. I never used Cursor, but Zed is my favorite editor
492.
by SwellJoe 4mo ago
Most successful scam in history. In awe at the gullibility of the entirety of investment/business media. (And, concerned for the sake of working class folks retirement accounts that are bound to get sucked up into Musk's maelstrom
493.
by SwellJoe 4mo ago
"It’s not beautiful but that’s OK." Really? I find his code elegant and concise.
494.
by SwellJoe 4mo ago
I was actually thinking of "Never attempt to teach a pig to sing; it wastes your time and annoys the pig." So, yes, mixed up idioms. But, the machine doesn't like arguing either. It is incapable of liking or disliking things.
495.
by SwellJoe 4mo ago
Friend, I'm not the one expending thousands of words ranting about some random woman that works at Anthropic.
496.
by SwellJoe 4mo ago
> If Claude is writing out combative and argumentative responses that's enough to call it "an argument". That also sounds crazy. I've never seen it become combative or argumentative. It is just a bland sort of polite
497.
by SwellJoe 4mo ago
You're misunderstanding what these models do. It is a limitation of LLMs. They don't have memory, they do not learn, they cannot learn. The sooner you let go of your desire to have them learn or remember anything, the sooner you w
498.
by SwellJoe 4mo ago
4.7 and 4.8 perform better than 4.6, so why is someone ranting about it being killed? And, Anthropic has 2500 employees, several of whom are higher up on the corporate hierarchy than "the woman who killed Claude". If someone is to
499.
by SwellJoe 4mo ago
"If you win an argument" Let me stop you right there. I am not arguing with a machine. You sound like a crazy person, when you say you are winning an argument with Claude. Claude is not my friend, I don't need it to agree wit
500.
by SwellJoe 4mo ago
Yeah, I see it write stuff to memory pretty regularly, maybe it works sometimes, but for things I want it to stop doing or always do, I make it impossible to do otherwise via lint or some style enforcement, or via a test that fails if code
501.
by SwellJoe 4mo ago
That doesn't follow. I can't google search, "are there any vulnerabilities in auth.go in this directory?" I can ask an LLM. And, if they find something, I can review it, and fix it, thus making infrastructure more resili
502.
by SwellJoe 4mo ago
Opus in recent versions is fine beyond 100k, but I usually do try to keep it under 200k. But, this is also why so-called "memory" systems are usually a mistake that make the models dumber. They don't have memory, they only ha
503.
by SwellJoe 4mo ago
Sorry, it's just on my mind, as I bought a Mac recently because it's the best hardware, but boy does the software suck.
504.
by SwellJoe 4mo ago
The performance would be abysmal spread across four Sparks, I'd think, though I guess MoE mitigates that somewhat. Still better to just pay for it in the cloud. (Though I've spent about $4k on local compute for AI experimentation,
505.
by SwellJoe 4mo ago
I did, but I prefer C. And, I prefer vim to an IDE.
506.
by SwellJoe 4mo ago
A lot of my earliest programming experiences were with Pascal. Apple Pascal in high school on Apple IIe and II+ machines. Later, Turbo Pascal on my dad's PC. I worked with the developer of IBM's Oberon system for OS/2 somethi
507.
by SwellJoe 4mo ago
Most of the agentic coding capable ones can. Even very small ones like Qwen 3.6 or Gemma 4 are surprisingly good at it.
508.
by SwellJoe 5mo ago
The biggest tech titans lined up to kiss the ring (and line Trump's pockets), and now we're seeing the obvious result. Those who bribe Trump and do exactly his bidding (including helping out with war crimes and surveillance of US
509.
by SwellJoe 5mo ago
Ah, that explains it. I guess I'll see it show up as separate usage pretty soon. Or, I'll find an alternative for that role (though Opus is clearly better than DeepSeek or MiMo or any other cheaper large model I've tried).
510.
by SwellJoe 5mo ago
I literally never hit usage limits in Claude Code on the $100 plan, and I feel like I'm using it as heavily as I can while still producing useful software. I could certainly make thinky machine go burr by giving it more busy work, but

More ›