mips_avatar

5 ms·

31.
by mips_avatar 3mo ago
I’ve found it helps a lot with reconciliation tasks where tools like openrefine can’t handle it. Like I wanted to tag blog posts with links to Wikipedia articles that are relevant. But the thing is whats relevant changes a lot based on cont
32.
by mips_avatar 3mo ago
The coolest project I’ve got this running on is improving the depicts metadata for photos on Wikipedia. A lot of times they won’t have the landmarks tagged correctly in a photo. So I will load in all the metadata that exists from each photo
33.
by mips_avatar 3mo ago
Qwen35ba3b can do a huge amount of data cleaning work on pretty modest hardware. Already have run about 100 billion tokens on it using 2x3090 gpus.
34.
by mips_avatar 3mo ago
Well yeah Claude caching is frustrating. If implemented correctly subagents are really cheap.
35.
by mips_avatar 3mo ago
One challenge/opportunity I've had is harnessing really wide running cheap agents. Any thoughts on how to move really cheap agents beyond basic summarization so we can go broader than the pricing of frontier llms allows?
36.
by mips_avatar 3mo ago
even 8x rtx pro 6000 is only 768GB of VRAM. IDK how anyone is going to run k3
37.
by mips_avatar 3mo ago
The more important question than subsidy is what is the tokenomics of running the model. If it's inefficient to run on an nvl72 cluster (or whatever the heck has enough vram to run a 3T parameter model), and k3 isn't very token ef
38.
by mips_avatar 3mo ago
It would be really interesting to redo the public benchmarks for kimi k3 but token normalize the costs. Ok so maybe k3 beats fable on terminal bench, but how many tokens did it use?
39.
by mips_avatar 3mo ago
I'm a creative person so my brain requires that I make something every day. Sometimes I make stuff that isn't very good. I've been told a lot by former bosses and random people that my desire to build stuff is frivolous. I c
40.
by mips_avatar 3mo ago
I think we need some law that if you are above a certain scale you have to publish traffic data as a gtfs feed, ie basically apple and google have to.
41.
by mips_avatar 3mo ago
The shared prompts are all cached so it's a cache read which is like 10x cheaper than a regular prefill
42.
by mips_avatar 3mo ago
The problem is nobody can build a real business on models they don't control. Cursor focused on making a great AI coding experience but they didn't control Claude and got destroyed once Anthropic started explicitly training Claud
43.
by mips_avatar 3mo ago
Thanks for making common crawl as good as it is. It’s a really important part of making the Internet better
44.
by mips_avatar 3mo ago
I kind of wish the recent Google monopoly court ruling had forced Google to open up their index to anyone, not just Perplexity/other big players.
45.
by mips_avatar 3mo ago
Qwen 35ba3b is ok, but it uses like 3x as many tokens as gemini 2.5 flash so you end up paying about the same to do an agentic run with gemini 2.5 flash and qwen 35ba3b.
46.
by mips_avatar 3mo ago
I find that Gemini flash 2.5 performs about as well as Claude sonnet for non coding agentic flows except it’s actually fast enough
47.
by mips_avatar 3mo ago
Yeah. Though I guess the point I thought of was like a deals site. That would have infinite pages and content
48.
by mips_avatar 3mo ago
that's hard to do with rendered content, oftentimes the result depends on a backend service. Maybe you should make the service it's running public but that might be a line most aren't willing to cross.
49.
by mips_avatar 3mo ago
It's such a good model for the price, for a lot of tasks it outperforms gpt5 at 3x the speed and 1/5 the price. The price jump from 2.5->3->3.5 has been so high.
50.
by mips_avatar 3mo ago
I feel like the solution is a better common crawl. As nice as it would be to block the frontier AI labs from getting access to information, we should reset the baseline of information accessibility so there's less marginal advantage on
51.
by mips_avatar 3mo ago
I’m uncomfortable with how much focus ai labs have focused on replacing workers. It’s easy to tune a model on an HR job description but a lot harder to enable people to be more capable.
52.
The end of consumer AI winter(jonready.com)
2 points by mips_avatar 3mo ago|1 comments
53.
by mips_avatar 3mo ago
Maybe micron/samsung/sk won't expand enough, but trust me the Chinese will. It might mean we don't get access to HBM for a while, but you can do a lot with infinite cheap DDR5
54.
by mips_avatar 3mo ago
The marginal cost of making RAM is zero and it's clear to everyone that there's near infinite demand, so the fab capacity is coming. Maybe the high end won't expand fast enough, but you can do a lot with Chinese ddr5.
55.
by mips_avatar 3mo ago
Yeah but Samsung/Micron/SK hynix are building new fabs, and Chinese Ram will be able to fully supply the lower end ddr5 ram market within 2 years. The crunch is temporary and on the other side of it there will be incredible hardw
56.
by mips_avatar 3mo ago
I know there's a lot of reasons to think that everyone will just use AI inference in the cloud, but I think if everyone had access to a dgx gb400 class machine with 512gb of hbm4 vram and 1tb of lpddr8x a lot of people are going to be
57.
by mips_avatar 3mo ago
I recently had to turn off posthog on my app, it was collecting so much information that wasn't needed that it was making my app unusably slow. I'm sure i'm missing some knob, but the fact that after an hour long claude code
58.
by mips_avatar 3mo ago
Yeah the problem is what's considered an adapted database. If it was strict it would mean apps like Alltrails (which is 90% openstreetmaps data) would need to list their trailmaps as open databases, but they don't.
59.
by mips_avatar 3mo ago
If it hadn't been for Starlink I wouldn't have been able to work remotely from my parents house during covid.
60.
by mips_avatar 3mo ago
I made a mistake, there is a $5k config with high memory bandwidth. The Max chip has two tiers (I incorrectly thought the tiers were based on memory capacity), you need the higher tier Max GPU upgrade (+$300) to get the 614 GB/s memor

More ›