potatolicious

6 ms·

31.
by potatolicious 1y ago
> "A: Using our own kernel helps us get rid of the baggage of legacy codes, bring the most optimal performance on our target hardware (XR/AR/VR) and achieve more efficiency than what we would've achieved on an existi
32.
by potatolicious 1y ago
I mean, there are so many factors. The piece covers many but I think there really are a few more that I've personally noticed: - Companies have generally tightened belts on discretionary spending, which means fewer people are being pai
33.
by potatolicious 1y ago
> "I suspect part of this is rich people coming up with use cases." Yes. Having been in the room for some of these demos and pitches, this is absolutely where it's coming from. More accurately though, it's wealthy p
34.
by potatolicious 1y ago
You can literally just Google this, there are many sources. Literally the top of the Google results: [1] https://www.politico.com/news/2025/03/21/trump-boeing-stealt... [2] https://www.usnews.
35.
by potatolicious 1y ago
It is in NYC also, except it's entirely unenforced. We need a lot more red light cameras. The nominal regulations on automotive behavior is pretty sufficient throughout the US, the main problem is that in most parts of the country tr
36.
by potatolicious 1y ago
> "Why does having launched in other cities matter if the new city brings up things that none of the other launched cities do?" New requirements come up all the time in technology. The existence of a new requirement isn'
37.
by potatolicious 1y ago
> "because LLMs are not part of the probabilistic environment of the domain you're engineering; rather, you're injecting new probabilistic inputs into your system" You do this as a process engineer also. You don&#x2
38.
by potatolicious 1y ago
> "I think they're trying to say AI generated code-pushers are often getting fuzzy on speccing out the behavior guarantees of their own software." I agree, and I think that's the root of the years-long argument of w
39.
by potatolicious 1y ago
> "When you consider that classical engineers are responsible for the correctness of their work" Woah hang on, I think this betrays a severe misunderstanding of what engineers do. FWIW I was trained as a classical engineer (m
40.
by potatolicious 1y ago
> "Hey, guys, listen, I know that this just completely torched decades of best practices in your field, but if you can't show me progress in a fiscal year, I have to turn it down." I mean, this is basically how all R&amp
41.
by potatolicious 1y ago
They do. The docs are right here: https://developer.apple.com/documentation/foundationmodels/g... If you have the latest betas installed you can call those APIs right now. They also support fine-tuning on the mode
42.
by potatolicious 1y ago
And yet just about any intro-to-programming tutorial gets something running on your local machine, and local machine development continues to be the default for most people, even though devving on a cloud machine is eminently reasonable. &q
43.
by potatolicious 1y ago
I remember that part too, I just don't feel like any of it was particularly egregious. Disappointing sure, but I suppose I disagree that anything about it was untoward. I'm generally pretty hostile to companies being acquired and
44.
by potatolicious 1y ago
I'm a bit lost here, and to be fair my memory from that long ago is a bit hazy... but I was a initial Pebble kickstarter and also got whatever the metal one was called. They were totally fine products, I enjoyed using them. I am no lon
45.
by potatolicious 1y ago
> "Whether it's a naturally emergent phenomenon in LLMs or specifically a result of its corporate environment, I'd like to know the answer." I heavily suspect this is down to the RLHF step. The conversations the mod
46.
by potatolicious 1y ago
Yeah, a lot of the comments here are really just addressing cults writ large and opposed to why this one was particularly successful. A significant part of this is the intersection of the cult with money and status - this stuff really took
47.
by potatolicious 1y ago
The term that comes to mind, and one of my favorite concepts, is "progressive disclosure", which is a concept we really ought to be more mindful of. One of the perks of just-a-text-file-with-a-bunch-of-addons is that it enables pr
48.
by potatolicious 1y ago
NAND is very, very slow relative to RAM, so you'd pay a huge performance penalty there. But maybe more importantly my impression is that memory contents mutate pretty heavily during inference (you're not just storing the fixed wei
49.
by potatolicious 1y ago
Would anybody buy the hardware though? Sure, datacenters will get rid of the hardware - but only because it's no longer commercially profitable run them, presumably because compute demands have eclipsed their abilities. It's kind
50.
by potatolicious 1y ago
It can , but I remain deeply unconvinced that the sub-agent architecture works as well as advertised. The trick with any layering like this is that your end-to-end reliability is subagent_reliability * routing_agent_reliabilitty. Neither a
51.
by potatolicious 1y ago
This feels like honestly the biggest gain/difference. I work on things that do a lot of tool calling, and the model hallucinating fake tools is a huge problem. Worse, sometimes the model will hallucinate a response directly without eve
52.
by potatolicious 1y ago
It's a bit worse than a branding problem honestly, since there's legitimate overlap between products, because ultimately they're different expressions of the same underlying LLMs. I'm not sure if you ever got a good rund
53.
by potatolicious 1y ago
There's a sweet spot in terms of generalization. Yes, painstakingly writing out an object definition in English just so that the LLM can write it out in Java is a poor use of time. You want to give it more general tasks. But not too
54.
by potatolicious 1y ago
Eh, this seems like a take that reeks a bit of "everyone is stupid except me". I do know the answer to OP's question but that's because I pickle my brain in this stuff. It is legitimately confusing. The analogy to diff
55.
by potatolicious 1y ago
Sure, go try it and evaluate it rigorously end-to-end, over a sufficient number and variety of tools. For the purposes of the exercise, let's conservatively say, maybe ~2000 tools covering ~100 major verticals of use cases. Even that m
56.
by potatolicious 1y ago
- If you get into a car crash with your iPhone a ML model detects this and automatically calls emergency services. - If you are wearing an Apple Watch a ML model is constantly analyzing your heart rhythm and will alert you to (some types of
57.
by potatolicious 1y ago
> "They have all data but can’t seem to get an llm that can set an alarm and be a chatbot at the same time?" This is actually one of the hardest frontier problems. The "general purpose" assistant is one of the singu
58.
by potatolicious 1y ago
It's a super interesting scene, and was a great eye-opener into how the mega-wealthy side of the world goes. A few things that pop into my brain in no particular order: - At the high end, everyone's got a gimmick. You can't j
59.
by potatolicious 1y ago
> "Respectfully disagree. Which oil-rich Sheikh did you see who only has one supercar in their garage and not all of those you listed?" Except this isn't the main market for Ferrari, McLaren, Lambo, et al. There are mayb
60.
by potatolicious 1y ago
> "However, if you work at a Bugatti dealership, you don't need to convince the oil-rich Sheikh why he should buy the 2 million dollar car, he already wanted to buy it before he even stepped in." Respectfully disagreed.

More ›