acchow

10 ms·

121.
by acchow 2y ago
> It sounds like it's so expensive and the difference in usefulness is so lacking(?) The claimed hallucination rate is dropping from 61% to 37%. That's a "correct" rate increasing from 29% to 63%. Double the correct r
122.
by acchow 2y ago
It looks like they built a WASM runtime in the TS type system. Implement some more OS calls and you should be able to get a browser running. And then WASM on WASM on TS types. Lol
123.
by acchow 2y ago
YouTube profits aren't broken out separately. However, Google's quarterly and annual reports do give Youtube Ad revenues, which were $36bn in fiscal 2024. That Youtube is not profitable is quite the strong claim..... https:/
124.
by acchow 2y ago
Ok, 2 months. Remember, the new B200 have 2.2x the performance of xAI’s current H100 “hardware edge”. So it only takes an order half the size. Or you could order the old H100 instead and avoid the B200 shortage.
125.
by acchow 2y ago
And only get to use 20k context length before it OOMs
126.
by acchow 2y ago
> but a compute edge is hard to match. xAI bought hardware off the open market. Their compute edge could dissappear in a month if Google or Amazon wanted to raise their compute by a whole xAI
127.
by acchow 2y ago
Because Tk draws using native OS APIs. So you’d need to run the whole OS in the browser. Or you could implement a new backend for Tk to draw using some web APIs
128.
by acchow 2y ago
Upvote does not mean “agree”. It means “discuss”. HN predates Facebook’s “like” button. It’s unfortunate people have been increasingly using it as “like”.
129.
by acchow 2y ago
There are only 2 books I could not put down once I started: The Davinci Code and When Breath Becomes Air. The first I raced through as fast as I could. The second I savored like a fine wine.
130.
by acchow 2y ago
The assumption is that since there is already a neural network that “got there” (our brains), we should be able to achieve the same thing synthetically. We just need to figure out how to train that network.
131.
by acchow 2y ago
Love and purpose
132.
by acchow 2y ago
Have you tried https://bolt.diy ? It does what you describe
133.
by acchow 2y ago
But the SOTA models basically all suck today. If people don’t think they suck, definitely in 1 year they’ll look back and consider those older models unusably bad
134.
by acchow 2y ago
Thats 100x quarterly revenue, or 25x annual revenue.
135.
by acchow 2y ago
> even being “a little” behind is a gap that grows exponentially every day This theory has yet to be demonstrated. As yet, it seems open source just stays behind by about 6-10 months consistently.
136.
by acchow 2y ago
> so the fast acceleration does not consume much energy Also, acceleration has no effect on the energy if you’re trying to hit the same target speed (unless there is friction, in which case you’d use less energy with higher acceleration)
137.
by acchow 2y ago
> if you use typing outside these two scopes you'd probably find many troubles. - encoding invariants and define valid evolutions of the codebase - memory safety without a garbage collector (see Rust’s Affine type system)
138.
by acchow 2y ago
How do you respond to people that say the profit motive is what drives innovation, and in the long term we’d rather have better, novel treatments sooner
139.
by acchow 2y ago
> He’ll remain an adviser to the Sonos board through June and get paid $7,500 per month until then. He’ll be paid a cash severance of about $1.9 million, and his unvested shares in Sonos will vest. Incredible vesting schedule
140.
by acchow 2y ago
I'm interested in which speaker and amp those were. Also, the plastic box :)
141.
by acchow 2y ago
Fiber keeps you feeling satiated for longer. Also, it’s bulking effect reduces your caloric intake while consuming the meal since the feeling of fullness can be reached with fewer calories.
142.
by acchow 2y ago
> I can run Qwen-2.5-coder 14B on my M2 Max MacBook Pro with 32gb at ~16 tok/sec. At least in my circle, people are budget conscious Qwen 2.5 32B on openrouter is $0.16/million output tokens. At your 16 tokens per second, 1 mil
143.
by acchow 2y ago
There’s value ad with ads targeting. Otherwise we go back to the Television days when everyone watched the same ads
144.
by acchow 2y ago
> You might be surprised to learn that manufacturing is larger industry in the US than our tech industry, or that our domestic oil industry is even larger US oil and gas industry has $240bn in annual revenues. This is about comparable to
145.
by acchow 2y ago
In my experience, a software engineer wears all the hats nowadays. Even crossing into product design and user experience.
146.
by acchow 2y ago
Location of data, policies, cross-cutting concerns, latency/performance requirements, etc. “Much later” is probably an exaggeration. You get to types “soon after” the initial design.
147.
by acchow 2y ago
Partially mind reading, but mostly mind control. You didn't think of the Previa out of nowhere. They led you to that thought. For more consumption.
148.
by acchow 2y ago
The way we do LLMs now is that the program and the data are one and the same. The program mutates itself as it "executes". This is probably also how the brain works since there is no hard separation between "memory" neur
149.
by acchow 2y ago
2.24 billion olympic sized swimming pools
150.
by acchow 2y ago
If you’re referring to speed, then thunderbolt does include PCIe support, including direct memory access.

More ›