Satam

I disliked them a bit, but then they stopped flying to a certain destination. I quickly realized that the other airlines were 3x more expensive. I realized I actually cared about price much more than any possible extra leg room or other perks, and that their super cheap flights are quality by itself.

Because LLMs are based on the abstract ideas of neural nets from brains. Say what you wish, but some problems were completely unsolvable before we adopted this paradigm. On some level, we must've gotten some ideas close to the right ballpark.

I had the same confusion initially, interestingly chat GPT gets it:

So while wolfgang42 wasn't there when Ulbricht was actually arrested, their realization created a vivid mental image of the event unfolding in that space, which made the story feel more immersive.

In short: they were reading about an old event, but it happened to occur in the same spot they were sitting at that moment. Hope that clears it up!

Is there reason to believe that the extra voters would've helped Kamala instead of Donald?

What successful projects have you birthed and how do they compare?

So you're saying it's better to be like most people: to promise nothing and then deliver exactly that, perfectly on time.

How do you reconcile his "bullshit" with the fact his spaceships are saving astronauts from space?

Wow! I love this take. Somehow with all this evidence of COT helping out LLMs, I never thought about using it more myself. Sure, we kind of do it already but definitely not to the degree of LLMs, at least not usually. Maybe that's why writing is so often admired as a way to do great thinking - it enables longer chains of thoughts with less effort.

I think it's because OpenAI's leadership lacks good taste and talent. Realistically, they haven't shifted the needle with anything really interesting in 2 years now. They're using the inertia well but that's about it. Their model is not the best, the UI is not the best, and their pace of improvement is not great either.

Except last I heard it's getting merged with the hated React router and it's not really clear what that implies, no? Haven't used either, just reading.

OP here. Thank you! I think your singling out of truly complex apps like Notion makes the situation more intuitive to me.

To clarify this, I think it's reasonable that token prediction as a training objective could lead to AGI given the underlying model has the correct architecture. The question really is if the underlying architecture is good enough to capitalize on the training objective so as to result in superhuman intelligence.

For example, you'll have little luck achieving AGI with decision trees no matter what's their training objective.

Yeah, I don't get it. I used Vue because it was straightforward. The options interface was Vue. Now, if it keeps getting more complicated and also has two ways to go about things, and the docs are split into two as well... Why not just go for React that's complicated and more popular?

My experience with Superhuman has been terrible. Back in the day, I tried to sign up. Couldn't. Then they started allowing sign-ups, but there was a mandatory onboarding/sales call to get started...

Finally, some time ago managed to sign up in hopes it could help to manage multiple inboxes with a unified inbox for all of my accounts. Nope doesn't have that. Canceled my subscription immediately yet I kept receiving their spam for a while.

Overall, a very scammy vibe.

I think you're right that for a dog to live its best life it needs the ability to spend a lot of time outdoors with relative freedom. Our labrador had the chance to live like that in an excessively very large garden for his last 4 years. I'm glad he got that, I think it made his life much better. Looking back, when he stayed with us in an apartment he must've been depressed.

Maybe the limitations of the software arise from the fact that it's being sold at a price that's relatively too low? It sounds like it's very niche software with only a handful of users like you at most. With a small number of paying users, even if each one is paying thousands per month, the tool's team might be too underfunded for it to do much more than maintenance.

To do it in-house, you'll probably hire 2-3 engineers. You'll aim to have smart people who can self-manage, design, and who can intuit the business requirements. Your part-time duty will become to be the product's "CEO" of that whole thing.

So you'll end up paying at least €16000/mo (3x€6000) in salaries alone. Data access, storage and infrastructure will probably cost a bit too. How much are you spending now?

Nothing survives over a long enough time scale. So the US will fall from grace and fail too. The question really is, is that going to happen within our lifetimes and what effects will we see in the meantime?

Great point. I think that might be in line with the idea if we take it to its full conclusion. Any way it goes, expect change. The meta mistake is assuming the patterns we've seen in the last 10/50/100/1000 years/etc. will continue.

If one looks at the last 10 years, they assume the market only goes up. If one looks only at the last 50 years, one assumes the US will continue to dominate.

I guess then, what is the mistake if we look only at the last 1000 or 10000 years? Maybe assuming the continued dominance of humans?

I think one should look at this as a nudge to think and re-examine our intuitions. The world and its 'order' has changed many times in the past. The leading empires rise and fall every couple of centuries, pandemics and wars devastate the world periodically. That's the pattern. Can we safely assume this time's the exception?

As a fun example, my understanding was that gold has always been the preferred standard for currency. Turns out silver was the far more prevalent standard for almost 5000 years. The switch to gold seems to largely have been caused by Isaac Newton's (yes, that one's) mistake in setting the exchange rate when he was the head of Britain's treasury. How many people don't know this yet exclaim how gold has always been the greatest form of currency?

I needed osm data at one point. Never managed to figure out how to do it the proper way. To get data you need, you need to download massive 100Gb files, in obscure formats, and use obscure libraries. Info is scattered, there are HTTP APIs but they’re limited or rate-limited and it’s not clear if you’re supposed to use them.

I know I’m ignorant and I’m happy the project exists, but the usability in the era where devs expect streamlined APIs is not great.

I ended up using some free project that had pre-transformed osm data for what i needed.

There’s probably lack of cpabalities on multiple fronts. RAG might have the right general idea but currently the retrieval seems to be too seperated from the model itself. I don’t know how our brains do it, but retrieval looks to be more integrated there.

Models currently also have no way to update themselves with new info besides us putting data into their context window. They don’t learn after the initial training. It seems if they could, say, read documentation and internalize it, the need for RAG or even large context windows would decrease. Humans somehow are able to build understanding of extensive topics with what feels to be a much shorter context-window.

RAG feels hacky to me. We’re coming up with these pseudo-technical solutions to help but really they should be solved at the level of the model by researchers. Until this is solved natively, the attempts will be hacky duct-taped solutions.

Tell 'em!

So far OpenAI's template is: amazing demos create hype -> reality turns out to be underwhelming.

Sora is not yet released and not clear when it will be. Dall-e is worse than mid-journey in most cases. GPT-4 has either gotten worse or stayed the same. Vision is not really usable for anything practical. Voice is cool but not that useful, especially with lack of strong reasoning from the base model.

Is this sandbagging or is the progress slower than what they're broadcasting?

That's only really true on the surface. So far the template is: amazing demos create hype -> once public it turns out to be underwhelming.

Sora is not yet released and not clear when it will be. Dall-e is worse than mid-journey in most cases. GPT-4 has either gotten worse or stayed the same. GPT-4 vision is not really usable for anything practical. Voice is cool but not that useful, especially with lack of strong reasoning from the base model.

Doesn't really matter if I do because I don't doubt some people feel strong emotions on this matter. But the fact remains, this is an ad and guitars aren't going extinct. People should get a handle on their emotions.

Also, I'm aware, that's why I wrote, "the way Devin is presenting itself". Apple doesn't have a factory where they're constantly crushing pianos either. So check your tone.

Easy fix: https://imgur.com/a/Ahwt6tr (although not sure which one is actually better)

If they apologized it's because it's the best PR move. The execs definitely aren't sweating over "destruction of human experience".

I understand we do these things because we find enjoyment in them. But none of these instruments and gadgets are even going extinct. There might be real shifts happening in our culture but at the end of the day this is just an ad. The emotional baggage that causes someone to be hurt by this should probably be handled at a personal level.

And thanks for sharing your thoughts. There isn't anything you said I'd disagree with. My original point was quite simple: people shouldn't be so soft. Although, I guess that's not really helpful and me saying that won't flip a switch in someone's head.

It was a rhetorical question. I can't force people to feel a certain way but I can certainly say that they're being overly dramatic.

And right, exactly, none of these things are actually extinct yet, so why dramatize? And if at some point they die off, it won't be Apple that caused it. It will happen because people stop caring, practising and using the things in question.