jbellis

5 ms·

61.
by jbellis 7mo ago
The explanation is terrible, but it's clear that it's not actually lossless.
62.
by jbellis 7mo ago
I'm interested if you're GPU-native and actually fast with my a7 IV raw images. Edit: oh wow, this is much older than I thought. Never mind. :)
63.
by jbellis 7mo ago
Benchmarking these now. Preregistering my predictions: Mini: better than Haiku but not as good as Flash 3, especially at reasoning=none. Nano: worse than Flash 3 Lite. Probably better than Qwen 3.5 27b.
64.
by jbellis 7mo ago
Now you're just being contrarian for the sake of being contrarian. https://www.arts.gov/sites/default/files/SPPA_Comprehensive_...
65.
by jbellis 7mo ago
The 90s aren't coming back to publishing. The audience who reads multiple books a month is going the way of the classical symphony attendee.
66.
by jbellis 7mo ago
You can, until they turn it off. Anthropic is pulling the plug on Haiku 3 in a couple months, and they haven't released anything in that price range to replace it.
67.
by jbellis 7mo ago
Your limitation after prefill is memory bandwidth. A maxed out Studio has less than a single 3090 (really).
68.
by jbellis 7mo ago
K2.5 isn't remotely a local model
69.
by jbellis 7mo ago
As far as benchmarketing goes they clearly went with prefill because it's much easier for apple to improve prefill numbers (flops-dominated) than decode (bandwidth-dominated, at least for local inference); M5 unified memory bandwidth i
70.
by jbellis 7mo ago
I chased down what the "4x faster at AI tasks" was measuring: > Testing conducted by Apple in January 2026 using preproduction 13-inch and 15-inch MacBook Air systems with Apple M5, 10-core CPU, 10-core GPU, 32GB of unified mem
71.
by jbellis 8mo ago
I bought an iPad for the same reason ~everyone does, for media consumption. But if I could use the hardware to do more interesting things then I'd be willing to spend more on a more powerful model.
72.
by jbellis 8mo ago
When the M1 ipad came out I said I'd upgrade from whatever my model year 2020 ipad is once I could run a Linux VM on it without rooting it. Still waiting.
73.
by jbellis 8mo ago
yes. in a similar vein, we're seeing that get standardized in coding agents as "don't have the agent use tools directly, have the agent write code to call the tools"
74.
by jbellis 8mo ago
are non programmers actually using openclaw successfully? because even "step 1 install your API keys" requires navigating concepts that are foreign to most "civilians"
75.
by jbellis 8mo ago
this is bullshit with a kernel of truth. none of the qwen 3.5 models are anywhere near sonnet 4.5 class, not even the largest 397b. BUT 27b is the smartest local-sized model in the world by a wide wide margin. (35b is shit. fast shit, but s
76.
by jbellis 8mo ago
TIL gemini still supports completion mode, that's super useful!
77.
by jbellis 8mo ago
He's talking about taking the government to court to force it to follow the law, not "maybe we'll get sued later."
78.
by jbellis 8mo ago
You just reinvented Skills
79.
by jbellis 8mo ago
yeah, g3p is as smart or smarter as the other flagships but it's just not reliable enough, it will go into "thinking loops" and burn 10s of 1000s of tokens repeating itself. https://blog.brokk.ai/gemini-3-pro-
80.
by jbellis 8mo ago
still no minimal reasoning in G3.1P :( (this is why Opus 4.6 is worth the price -- turning off thinking makes it 3x-5x faster but it loses only a small amount of intelligence. nobody else has figured that out yet)
81.
by jbellis 8mo ago
really too bad that the codex models are so tightly coupled to the codex harness as to be useless for everything else edit: not useless in a absolute sense, but worse than the vanilla gpt models
82.
by jbellis 8mo ago
M2 was one of the most benchmaxxed models we've seen. Huge gap between SWE-B results and tasks it hasn't been trained on. We'll put 2.5 on the list. https://brokk.ai/power-ranking
83.
by jbellis 8mo ago
Yes, very similar results here ( http://brokk.ai ) We got lines-with-anchors working fine as a replacement strategy, the problem was that when you don't make the model echo what it's replacing, it's literally dumb
84.
Syntax-aware diffs without the false postives(arxiv.org)
2 points by jbellis 9mo ago|0 comments
85.
by jbellis 9mo ago
Good to see that Anthropic is honest and open enough to publish a result with a mostly negative headline. > Importantly, using AI assistance didn’t guarantee a lower score. How someone used AI influenced how much information they retaine
86.
by jbellis 9mo ago
Not because of a sudden outbreak of sanity, but because they have CT scanners now.
87.
Open Weights Coding Models in 2025(blog.brokk.ai)
3 points by jbellis 10mo ago|0 comments
88.
by jbellis 10mo ago
Love to see people leveraging static analysis for AI agents. Similar to what we're doing in Brokk but we're more tightly coupled to our own harness. ( https://brokk.ai/ ) Would love to compare notes; if you're
89.
by jbellis 10mo ago
Now I'm curious about what fastutil's implementation is doing.
90.
by jbellis 10mo ago
> where are the professional tools, meant to be used for people who don't want to do vibe-coding, but be heavily assisted by LLMs? This is what we're building at Brokk: https://brokk.ai/ Quick intro: https:&#x

More ›