HNHacker News
TopNewBestAskShowJobs

loveparade

2,076 karma · joined December 30, 2022

Meet Me At The Love Parade
submissionscomments
loveparade··on Sam Altman, Greg Brockman and others to join Microsoft
Over time, the new team at Microsoft will slowly drain the OpenAI talent as MS will provider fewer and fewer resources to OAI and prioritize their own team they have full ownership owner. OAI will die a slow death due to the talent and resource drain. In a year or two, Sam and Greg will end up with the same people at MS that are now still at OAI. Okay, perhaps minus Sutskever.
loveparade··on Emmett Shear becomes interim OpenAI CEO as Altman talks break down
I think there's an important difference between not understanding and not doing the work yourself because you are in a managerial/advisor position. If you give Hinton a recent paper on LLM RLHF he will understand the nuances in it, he just delegates the actual work. If you give Emmett Shear or whoever such a thing, they almost certainly don't. For a deep tech company focused on research (not some consumer SaaS thing) I don't think you can be a good CEO if you don't even have an understanding of what you are building.
loveparade··on Emmett Shear becomes interim OpenAI CEO as Altman talks break down
Maybe read my comment again? There is no lie. I said accepted, not attended. And if you've ever tried to get into a PhD program at Stanford you'd know that isn't easy unless you have, you know, at least a little bit of knowledge of the field. So the parent claiming that Elon Musk doesn't even know basic Physics is blatantly wrong.
loveparade··on Emmett Shear becomes interim OpenAI CEO as Altman talks break down
You can hate Elon Musk as much as you want, but he absolutely knows a decent amount of Physics and Material Science. That's what he studied and has always wanted to do before PayPal. He even get into a Material Science PhD at Stanford. He drove a lot of the engineering at both Tesla and SpaceX. And without that, probably none of these companies would've worked out.
loveparade··on Emmett Shear becomes interim OpenAI CEO as Altman talks break down
For a change, maybe it'd be a good idea to get a CEO who at least has faint idea about how ML works on a technical level? Would you like a company that builds rockets to be led by some salesman who doesn't understand basic Physics? Probably not. You don't need to be Hinton or Schmidhuber, but come on, who thinks it's a good idea to have these typical "Silicon Valley VC CEOs" lead an AI company?
loveparade··on Ask HN: Does Hacker News serve you?
The most useful aspect is the curation. HN does a pretty good job at curating headlines so that I can save time and get the most important news items each day without visiting other social media sites, most of which push monetization and recommendations in annoying ways.

As for reading the comments, it's just entertainment and the same as any other social media site.

loveparade··on Greg (2017)
Google and Meta have less compute infrastructure purpose-built for training and testing LLMs than OpenAI does. It's something that OpenAI has been investing in for several years, while Google/Meta/etc have just started playing catch up very recently.
loveparade··on Greg (2017)
You're massively overestimating the skills needed to build "competitive models". Researchers like Ilya were essential several years ago when OpenAI got started, because at that time nobody even knew what to build or how to go about it. It was all research, and OpenAI was solely focused on RL. But these days, LLM code is 95%+ data and infrastructure engineering and very little novel research. Even things like LLM speedups via new attention mechanisms are more engineering than they are research.

Above all else, what you need to build competitive models is a huge amount of money and access to compute. Money alone doesn't solve that either due to the global hardware shortage. You are competing for a globally limited pool of hardware with all other companies.

loveparade··on [dead]
What does that even mean, "make it"? AGI that nobody can even define? Obviously not, nobody is gonna make it.
loveparade··on Ask HN: Have you seen any useful GPTs yet?
Like you said, it's mostly influencers trying to cash in on the hype and I don't expect that to change.

> Aren't these essentially pre-configured prompts?

They also include custom knowledge and actions, and I don't think these are simply added to the prompt. But I don't think OpenAI has published how this data is fed into the model internally?

In general, I've had mostly bad experiences with models that include retrieval or external APIs. It's just too brittle because the model needs a very precise understanding to retrieve what you want, not fuzzy language. The by far best experiences is when you can put everything into the prompt.

loveparade··on Ex-Kotaku staff go independent and launch Aftermath
Player reviews for games aren't much better, unfortunately. So many games get review bombed for political reasons. It's the Amazon equivalent of "this book sucks because my package was late" but orders of magnitude more frequent. It's nearly impossible to find high quality game reviews.
loveparade··on Charm has raised $6M in funding
And with that the usual enshittification brought about by VC funding and incentives will begin. In 2-3 years it'll be unusable. Congrats to the founders though, have a nice exit down the road.
loveparade··on Current thoughts on social media
Well, technically speaking it's impossible for them to "find you" as there is no way to distinguish between a sophisticated enough bot and a human user.
loveparade··on Current thoughts on social media
> It sucks that these platforms have made creating a single tool or site for interacting with each other increasingly impossible

Is it really impossible or just very difficult? To me it seems like it's the latter since alternative frontends as in libredirect [0] exist for many sites. You can't rely on APIs and would need to build custom browser simulation for each site. That's certainly difficult, but doesn't seem impossible?

I don't see how these sites could ever make automation completely impossible without locking out humans as well.

[0] https://libredirect.github.io/index.html

loveparade··on Sponge cleans all teeth perfectly in just 15 seconds
Would love to hear real reviews of this from people who've tried it. If it really works as well as advertised that'd be amazing.
loveparade··on As the public begins to believe Google isn’t as useful, what happens to SEO?
It's not yesterday, it was 20 years ago when Google excelled at exact keyword matches. It has been a constant evolution away from that.

I hate the new Google as much as others, but if you don't adapt your search habits for 20 years when the whole ecosystem around you has been obviously changing, that's kind of on you. Just use another search engine that fits your use case. Personally I use Kagi and I haven't touched Google for the last year at all.

loveparade··on As the public begins to believe Google isn’t as useful, what happens to SEO?
Yes, you're missing the point of the poster above you. He's saying that the kind of technical searches you are trying to do are irrelevant to 99.99% of internet users and Google is optimized to give superficial but relevant-looking results to those kinds of people. For that metric Google does "work well"

So it's not that Google doesn't work, it's that you're using the wrong tool for your use case. If you want highly technical content or precise keyword search you shouldn't use Google. It's like you're going on Tinder to find a marriage partner.

loveparade··on As the public begins to believe Google isn’t as useful, what happens to SEO?
I disagree, but it depends on your definition of "working well". Yes, Google works better for my mom these days if you measure whether she is happy with the results she gets. However, the results she gets are mostly misinformation or spam. She just doesn't realize that's the case and is happy with the results regardless.

Measuring whether or not results are relevant to the query is the wrong metric to use. You don't want highly relevant misinformation. You want good information.

loveparade··on 'It's quite soul-destroying': how we fell out of love with dating apps
I mean, it's pretty obvious isn't it? The ratio on most dating apps is 10-100x more men then women, a large fraction of fake profiles and scams, and the companies pushing predatory monetization hacks as a result of all that.
loveparade··on Ask HN: Will you gain Bitcoin exposure if a Bitcoin ETF is approved?
> Is it just a vehicle to trade Bitcoin (as in a single cryptocurrency; BTC) on a normal stock exchange?

Yes

> Is it an index-weighted portfolio of BTC and other (e.g. Ethereum)?

No

> Does the fund actually own BTC or is it technically a derivative like a synthetic (regular) ETF?

Yes, a spot ETF is backed by "physical" Bitcoin

loveparade··on FreeTube – The Private YouTube Client
It's not about the price. The whole UX is bad with their pushing of irrelevant clickbait content.
loveparade··on Google’s dominance under siege: Antitrust trial threatens sweeping changes
YouTube
loveparade··on Show HN: Firefox add-on to open YouTube videos in alternative front ends
Thanks, I wasn't aware of Unhook. I know the I can use the subscriptions feed, but it doesn't solve all problems. When I open YT on my phone, or on my TV, I still get blasted with recommendations and clickbait. When I search for a keyword I get relevant results interspersed with clickbaity content (similar to Google SEO spam), and so on. This problem permeates the whole YT experience because the business model is maximize engagement metrics and ad view time at all costs.

Unhook looks good indeed, I guess I am just a bit frustrated that I need external extensions, many of which are linked in this thread, and hacks to fight against services that try to monetize my attention. Especially when I already pay for them.

loveparade··on Ask HN: How do I stop wasting time?
I believe that everyone has experienced this, but the reasons behind it, and what techniques may work, tend to be different for everyone. A lot of online content is very addictive.

You may want to investigate if there is a deeper reason behind your avoidance for studying. For example, uncertainty around "is this stuff actually useful" or "will I succeed" may create anxiety and avoidance and push you towards distractions, whatever they may be. You could then work on accepting or eliminating that uncertainty.

On the other hand, maybe it's much simpler and you're just addicted to some of the junk you find online, similar to how some people are addicted to porn.

loveparade··on Websites of 2000 [video]
Wow, this makes me sad. Back then, websites on the internet were merely a useful tool that you controlled. Now it's the opposite. Websites on the internet control you by completely hijacking your attention with highly optimized clickbait and ads that humans are wired to engage with.
loveparade··on Show HN: Firefox add-on to open YouTube videos in alternative front ends
I don't care as much about the privacy as I care about not messing up my Youtube recommendations. Any time I accidentally click on or hover over some BS video my front page is full of clickbait spam and I need to manually clean up my watch history.

I guess (?) it works for monetization, but the UX of YouTube and their pushing of clickbait content has become horrible over the past few years. I just wish they stopped trying to be "smart" about what I want to watch. I'm already paying for Premium, just give me an option to make Youtube stupid.

loveparade··on Ask HN: Name a non-cryptocurrency blockchain project in use in the real world?
I totally agree with this take. I like to describe Ethereum as a "platform for scams" because its by far biggest use case are scams. Scams are just so easy when you have money built-in at the base layer. Just imagine a world where decades ago SMTP had launched with built-in payments. Leaving aside that adoption would've been impossible during that time because people are scared of putting their credit card info online, it likely would've resulted in SMTP to fail and being used almost exclusively for scams since it would've been so easy to extract money from people via email. Crypto has that problem.

That doesn't make SMTP (or Ethereum) itself a scam. But it's also undeniable that a platform for scams doesn't have a positive impact on the world.

loveparade··on Ask HN: Name a non-cryptocurrency blockchain project in use in the real world?
> Ok but I do understand it, I also understand how cool the concept of the blockchain is _but_ real application of the blockchain outside of crypto is practically non-existent or could be better served by not using a blockchain.

I don't disagree with you here, I think the by far biggest use case is money/currency. I'm just commenting on how people here hate on anything related to crypto without any understanding.

> As for snarky comments about “I thought crypto was decentralized”. Yes it’s snarky but it’s a totally valid criticism.

I disagree here. It's like saying "SMTP sucks because I got scammed by an email from a Nigerian prince" - The criticism that wallets are hard to use is valid, but people on HN commenting on these threads don't even understand that. They just hate crypto for whatever reason without any understanding.

loveparade··on Ask HN: Name a non-cryptocurrency blockchain project in use in the real world?
> I can’t imagine why HN thinks crypto related stuff is bullshit /s

Because the vast majority of people doesn't understand anything about it. And most don't want to because they've already made up their mind. You just have to look at any thread about crypto and it's full of misinformation and idiotic takes from people who obviously have no idea what they're talking about. E.g. threads on some kind of centralized exchange scam that has with comments like "I thought crypto was decentralized" or "I thought blockchain solved this" - Everyone likes to have an opinion on crypto even when they're completely uninformed. Of course it doesn't help that 99% of crypto projects are scams, but that doesn't make the underlying protocol a scam.

AI doesn't really have that problem. You don't need to understand anything about how models work to use an LLM. AI also doesn't have payments built-in at the protocol layer, so it's much harder to scam people with AI. But crypto is fundamentally different. It doesn't have easy-to-understand apps that make you more productive like an LLM does. It's highly political and ideological and operates on a layer (banking) that most people have never had exposure to and don't think about in their daily lives.

> burning a ton of power for dubious results

You realize that pretty much all crypto projects are built on top of Ethereum or Ethereum L2s which barely burn any electricity, right? There are (nearly) no projects built on Bitcoin because well, you can't really build anything on Bitcoin.

loveparade··on Playing Pokemon Red with Reinforcement Learning
Pokemon Red isn't a huge game if you have access to the state (not just pixels) and prune it properly. The RL algorithm used here is so sample inefficient that it's pretty much just a stupid search over the space. You can likely solve the game easier using traditional search algorithms. This shows how to use the wrong tool for the job. It would've been more interesting to use algorithms optimized for this kind of problem, not some off-the-shelf PPO that's just inappropriate.
← PreviousPage 8 of 13Next →