mrbungie

22 ms·

391.
by mrbungie 2y ago
If you go down that route probably you will end up designing websites that are also anti accesibility / anti screen readers / etc. I don't see a way around it, without completely redesigning the protocols/languages/
392.
by mrbungie 2y ago
And if AGI eventually comes from OpenAI. I'm starting to doubt that.
393.
by mrbungie 2y ago
This. If you don't trust your provider when it says it won't use data you embed in your LLM API Calls, why do you trust them when you use any of their other services?
394.
by mrbungie 2y ago
They still will need a lot of materials for adapting those buildings for higher cooling and power needs.
395.
by mrbungie 2y ago
Surely concrete is going to be hell of a constraint for both compute (datacenters) and power for said compute (power plants). And also expect for higher water usage. Interesting times ahead.
396.
by mrbungie 2y ago
Subhuman intelligence/reasoning with superhuman knowledge access. Searle's chinese room argument is starting to make sense in this case.
397.
by mrbungie 2y ago
Actually good points. I just hope other players catch up.
398.
by mrbungie 2y ago
NVDA banned CUDA translations layers, that's pretty exclusionary because it denies interoperability. There was already a discussion about it here: https://news.ycombinator.com/item?id=39592689
399.
by mrbungie 2y ago
Well, then let's hope EU puts its eyes on NVDA and CUDA, another walled garden.
400.
by mrbungie 2y ago
> Replacing search is enough to warrant trillion dollar valuation for LLMs, question is just if they can get that good. I would add that it needs to be more than just good, it also needs to be cheaper for mass adoption. Spoiler: classic
401.
by mrbungie 2y ago
Demand driven mainly by hyperscalers. And hyperscalers (MSFT as an example) are aggressively forcing customers (using tactics such a latency/performance compromises, one example being completions in 30 secs vs 1.5 min for the same prom
402.
by mrbungie 2y ago
Gemini might not be as cool as GPT, but it already runs on TPUs. So the "G" in FAANG is not as dependent on Nvidia as the others.
403.
by mrbungie 2y ago
If it worries you, maybe open an issue? No sane man would allow a weird license that's an API call away from screwing up your own products.
404.
by mrbungie 2y ago
I think that is being transparent with what actually happens in the real world (engineers, at least in part, being human and emotional in their decisions), rather than just talking about impossible ideals (engineers thinking about tradeoffs
405.
by mrbungie 2y ago
Academic/theorical computer scientists, totally. But pragmatic programmers / software engineers, try reading the average PhD's python code. Usually it is a mess. Knuth may be an exception, rather than the rule.
406.
by mrbungie 2y ago
They should totally try to eventually compete with the Mac Mini, even if only via a clear product/vision roadmap. No one is going to develop for a dead platform.
407.
by mrbungie 2y ago
Pretty sure that license header ended up in the codebase from a clever guy making a PR or it was just a mistake.
408.
by mrbungie 2y ago
Do they get a free pass because of accelerationist reasons? If anything this is a reason to stall/brake. sama tweet after the demo + SJo's press release + OpenAI not even risking it and pulling out the voice from ChatGPT should ra
409.
by mrbungie 2y ago
Probably anyone but her inner circle can "approach her" directly. I would expect any other kind of connection to be made through her agent.
410.
by mrbungie 2y ago
The "Sky" voice and it's likeness to SJo's have been there in the ChatGPT app for months.
411.
by mrbungie 2y ago
You're not arguing your way out of jurisprudence, especially when the subject is a human and not a device nor IP. They (OpenAI) fucked up.
412.
by mrbungie 2y ago
Any non-trivial complexity codebase eventually implements a mediocre SQL/Lisp/etc.
413.
by mrbungie 2y ago
That's why these things should run code in protected sandboxes. Not to do it in a "protected mode" would be negligent.
414.
by mrbungie 2y ago
Don't hold your breath. Gemini 1.5 Pro/Ultra was meant to be the hottest shit ever, and we know how it ended up mere days later.
415.
by mrbungie 2y ago
Because you can charge up a lot more when adding hot and hyped features like LLMs instead of doing good engineering.
416.
by mrbungie 2y ago
Well, you have a point. I've always thought that Hacker News <> YCombinator, but maybe the truth is in the middle. At the very least, this is food for thought.
417.
by mrbungie 2y ago
The amount of copium in this response is astounding. Yes, there is a noticeable negative response from HN towards Google, and there has always been especially when speaking about their weird product management practices and incentives. Goog
418.
by mrbungie 2y ago
Probably there will more nuance than that. And doomscrolling is a thing, you know.
419.
by mrbungie 2y ago
If the advertisers still use A/B testing and WANT to learn from said tests. Sure. If the advertisers use multi-armed bandits you wouldn't care less about the segments themselves.
420.
by mrbungie 2y ago
That's the way of life. Older people think younger people are stupid and reckless, and viceversa. And the younglings think they "figured it out" like no one before them. But no one ever tried to understand each other in the p

More ›