buu700

32 ms·

271.
by buu700 1y ago
Because if the LLM is a black box and its output must ultimately be verified by humans, then you can't treat conversion of prompts into code as a simple build step as though an AI agent were just some sort of compiler. You still need t
272.
by buu700 1y ago
I assumed they were describing AI itself as a black box (contrasting it with deterministic code generation), not the output of AI.
273.
by buu700 1y ago
Overloading of the term "generate" is probably creating some confused ideas here. An LLM/agent is a lot more similar to a human in terms of its transformation of input into output than it is to a compiler or code generator. I
274.
by buu700 1y ago
16 years ago, when my friend and I had some downtime while working at a summer camp, we got bored and put together a crazy "AI"-assisted* essay that was filled with wild ranting, references to people at camp, and other camp inside
275.
by buu700 1y ago
I had the same thought, with the caveat that I'd be less specific than "barbell training". Adjustable dumbbells take up less space in a home gym, and with few exceptions and caveats, you can do all the same movements and prog
276.
by buu700 1y ago
I think there's a useful distinction that's often missed between AGI and artificial consciousness. We could conceivably have some version of AI that reliably performs any task you throw at it consistently with peak human capabilit
277.
by buu700 1y ago
I know a lot of less tech-literate users who would leave something like this straight away thinking they're being hacked or something daft. That reminds me of the time my friend and I put together a silly shell script called pentagon
278.
by buu700 1y ago
The fact that you can more info from an LLM than it holds is actually a pithy description of this whole challenge.
279.
by buu700 1y ago
I did something pretty similar with USPS around 15 years ago. Walked into the post office, handed them a banana, they slapped a label on it, and off it went. A few weeks later I heard from my friend in Monaco that her mom had gone to check
280.
by buu700 2y ago
I've been flagged as a potential shoplifter by the self-checkout at the grocery store based on some video analysis of CCTV footage of my hand motions. Shopping in 2025 must be a frustrating experience for magicians.
281.
by buu700 2y ago
* Lower corporate income tax to the global minimum of 15% * Reshoring incentives * Regulatory reform to remove barriers to building things * Strict export controls around AI, robotics, and fusion * Massive subsides for production of useful
282.
by buu700 2y ago
Agreed. It's better than no data, but I don't really see what a study like this can hope to prove one way or another. Giving a handful of people free money isn't UBI any more than the lottery is UBI. It doesn't provide a
283.
by buu700 2y ago
I like the idea of CVSS, but it's definitely less precise than I'd like as-is. e.g. I've found that most issues which I would normally think of as low-severity get bumped up to medium by CVSS just for being network-based atta
284.
by buu700 2y ago
Not all measures: https://www.marketwatch.com/story/this-chart-shows-the-drama... . It may not have declined in absolute terms, but it still hasn't kept pace with the rest of the economy.
285.
by buu700 2y ago
I was actually a regular user of GIMP for years before I learned that the word had any other definition. While working at a summer camp as a teenager, I happened to install it on an office computer. One day I'm doing some work in the o
286.
by buu700 2y ago
That's fair. I'd basically agree, which is why I decided not to hybridize SPHINCS+ with Dilithium as I did with the two post-quantum encryption algorithms. Having said that, while SPHINCS+ seems highly likely to be safe (particula
287.
by buu700 2y ago
In Cyph, I went with Kyber (lattice-based) combined with HQC (code-based) for encryption. NTRU Prime also may be a good option if Kyber is ever broken in a way that doesn't fundamentally break all lattice crypto. For signing, I treat D
288.
by buu700 2y ago
Really glad to see it. It sounds like their reasoning is essentially the same as mine for implementing HQC alongside Kyber in Cyph three years ago: having a different theoretical basis makes it highly complementary to Kyber/ML-KEM, but
289.
by buu700 2y ago
That framing of the issue makes a lot of sense. It might still be a reasonable middle ground if Google doesn't own Chrome itself, or control the proprietary bits or any related backend services, but still maintains leadership of Chromi
290.
by buu700 2y ago
It shouldn't be a war on anything. Because we insist on treating adults like children, we're taking money that would have gone to tightly regulated pharmaceutical corporations like Bayer[1] and funneling it into violent criminal s
291.
by buu700 2y ago
I had a similar gut reaction, but on reflection I think 4.5's is actually the better response. On one hand, the response from 4.5 seems pretty useless to me, and I can't imagine a situation in which I would personally find value i
292.
Intel principal engineer bemoans potential TSMC takeover, touts 18A advantage(tomshardware.com)
10 points by buu700 2y ago|0 comments
293.
by buu700 2y ago
It shouldn't be the tool's job to tell the user what is and isn't a good question. That would be like compilers saying no if they think your app idea is dumb, or screwdrivers refusing to be turned if they think you don't
294.
by buu700 2y ago
Like libsodium.js and pqcrypto.js. They're distributed as single self-contained JavaScript files, but under the hood have wasm embedded using the emscripten SINGLE_FILE flag with an asm.js fallback.
295.
by buu700 2y ago
I don't know offhand what TSLA's stock price is, but I think it's fair to say that the speculative value is at least very high if you believe that Tesla Optimus is at all a viable competitor to Boston Dynamics Atlas. I see mu
296.
by buu700 2y ago
Related: https://www.foxbusiness.com/politics/doge-could-create-mobil...
297.
by buu700 2y ago
That's fair. It isn't outside the realm of possibility that the Pope and every previous Pope is and was a true agent of God (despite how historical issues around papal succession and legitimacy may complicate that story), and ther
298.
by buu700 2y ago
I think it's a more interesting approach to disregard the metaphysical claims, which are inherently unfalsifiable and thus irrelevant to life in this universe, and look at religious texts as constitutions governing human behavior and m
299.
by buu700 2y ago
Same, after a certain point (for TypeScript particularly) I found it was too big a hassle not to switch to VS Code. All these years later, I still have VS Code configured to look and behave like Sublime. I found it funny one time when someo
300.
by buu700 2y ago
I would argue that both things are true: 1. The future of inference for ChatGPT-style direct consumer usage is on-device. Cloud-based inference is too gaping of a privacy hole in a world where some level of E2EE is rapidly becoming the defa

More ›