doghut

Professional AI-initiativist, prospective POTUS (2044), prospective CEO of the McDonald's Corporation, prospective CEO of United Airlines.

[doghut @doghutty]

Love that even @AnthropicAI's small ships are really cool; they've really been on a roll as of late. They know exactly what people would want, and they're delivering. Even something as cosmetic as this update far outweighs whatever @OpenAI did on that 28 day thing (what did we even get out of that?).

@DarioAmodei's model of "no vague posting, none of that bs, just purely shipping good models and good products and over-delivering" is really good. @thsottiaux can learn a thing or two from that.

Again, just to be clear, I'm not trying to hate or single out any specific entity, and if anything, I want both Anthropic and OpenAI to do well (as with Google DeepMind and MSL/Meta), because competition is a phenomenal thing, but OpenAI has absolutely been fumbling the bag as of late. @sama, we want more transparency, better products (models in particular), and less bs vague posting.

[ClaudeDevs @ClaudeDevs]

You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own.

Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!

[doghut @doghutty]

Agreed. I rather they be good so they serve as additional true competition, which benefits everyone, but distilling isn't gonna get them there. It also calls into question whether they really have true talent that's capable of building better models to begin with, or whether they were just fraudulent copy-cats to begin with. Of course the best AI talent in the world is largely concentrated within Anthropic, OpenAI, and Google DeepMind, but the performance from some of these open labs have just been so disastrous that I'm seriously doubting whether they have any true talent at all.

[doghut @doghutty]

88% of NVIDIA's employees want Jensen to stay as CEO (as per an https://t.co/14jtOQiHsV survey on Yahoo Finance). That's a massive figure, especially after a scandal, which clearly demonstrates that most would consider his personal wrongdoings to be independent of his leadership.

I'm not defending his affairs, but it's clear that Jensen has been (unfairly) the target of a media feeding frenzy, like with this incredibly nasty and bait-y headline here from Bloomberg (attached image) that they know very well will garner a significant amount of user engagement. "oOooooOo loOk eVeRyOne, jEnSen cAlLEd himSeLF a pIeCE oF sHiT plSsss cLick ouR aRtICle oh also you have to pay a subscription to read the article" (yes, Bloomberg actually does have the article behind a paywall).

People have done far worse than him with little to no coverage. And regardless of his personal affairs, I think most would agree he was a phenomenal founder & CEO. Jensen IS NVIDIA, and to remove him is a massive mistake. Had it not been for the smear campaign from the media this past week, I'm almost certain he would still be CEO right now.

I'm also a strong believer that he's one of the most well-aligned figures (by that, I mean well-aligned to the general public) in the world. I much rather have him lead the future of compute and AI as opposed to a lot of other powerful figures.

[doghut @doghutty]

i wonder if this is how people felt when Steve Jobs got fired in '85. what's even more interesting is if we'll see @JensenHuang potentially get rehired like Jobs as well

[doghut @doghutty]

At this point, I see absolutely no reason to have an OpenAI subscription (and I'm glad I cancelled several months back when I observed this decline). Anthropic is essentially better than them on all fronts right now, so there's really no justification in paying OpenAI at all. Just doesn't seem like they can do anything right at this point (even with dots here, which is supposedly their focus right now).

[doghut @doghutty]

@thsottiaux have you guys considered shipping something actually good? like, idk, a new SOTA model? just my two cents.

[doghut @doghutty]

im not surprised if elon told dario in private "yo ill give you guys all the compute you could ever want, just destroy openai for me" lol

[doghut @doghutty]

It's still insane to me how large the gap has gotten. @AnthropicAI leapfrogged everyone so substantially these past several months that I've found myself solely using their models for anything even remotely serious. Rumor has it (pretty much just based on vague posting from @thsottiaux and some other guys) that @OpenAI is releasing 6.2 Astra tomorrow, but at this point I've pretty much lost all faith in them. I'm almost confident that they're well aware they can't beat Anthropic at this rate, and are now solely aiming for 2nd place and the everyday consumer (casual user) market.

Ideally, I'd like to see OpenAI do well so they can serve as a legitimate competitor in the AI sphere, given that Anthropic and OpenAI are essentially the only true two leading frontier labs, but I honestly just don't see it happening at this point.

[doghut @doghutty]

I essentially never read prompting guides since, most of the time, it's just common sense. However, Fable 5.2 is (allegedly) different. I took a look through Anthropic's official prompting guide and it appears that there's two fairly significant quirks (that many users have already reported encountering), and Anthropic has fully addressed these in their prompting guide for Fable 5.2. https://t.co/4U6YEizlx7

1. It has a tendency to stop early (especially on lower effort levels like low and medium). To prevent this, one can explicitly tell Fable 5.2 to avoid stopping unless intervention from the user is absolutely necessary.
2. It's extra careful with codebases and avoids making changes (it inherently leans towards additive programming and avoids changing existing code) unless the user explicitly tells it that it owns the codebase and has full authorization to make any changes it sees fit. Basically, just tell Fable 5.2 that it owns the workspace and is pre-authorized to do anything within that workspace.

[doghut @doghutty]

Yeah, okay, Anthropic won. They are THE leader in AI, and at this point, it's pretty much indisputable. Credit where credit's due: Anthropic is phenomenal at what they do, and no one else comes even close.

I honestly think OpenAI may have just given up at this point. There's a reason they're pivoting to other more consumer-grade services like dots: they know they can't compete with Anthropic model-wise (anymore). Then again, they don't have to. They have ChatGPT, which is still the dominant player when it comes to the regular day-to-day consumer market. In contrast, Claude is the go-to for developers, enterprise, and essentially anyone who uses AI for more serious use cases.

Ultimately, my point is that there's a chance we'll never see OpenAI with better models than (or even on par with) Anthropic ever again. Over these past couple of years, we've gone from OpenAI being the dominant player, to Anthropic getting neck and neck with them, to OpenAI taking a slight lead again (with a significant gap closure this time around), to now Anthropic absolutely obliterating OpenAI on essentially all fronts.

Anthropic won. OpenAI (very likely) straight up quit, because they know they've lost. Now I want OpenAI to prove me wrong because competition is highly beneficial, and AI is an incredibly fast-moving sphere, so things can certainly change, but at least as of now, things are absolutely looking bleak for OpenAI.

[ClaudeDevs @ClaudeDevs]

We've given all Pro, Max, Ultra, and Team users one free full reset, across all models and limits. Use it anytime before December 10 at 2:00PM PT.

We recently fixed a few small issues that used more limits than intended, and the Fable resets didn't do much for those of you who'd already hit your weekly limits for all models. While these issues didn't affect everyone, we want to make up for the trouble.

You'll find it in Settings > Usage, or at https://t.co/5YRzheAZyQ.

[doghut @doghutty]

I've really missed Fable, and I'm glad @AnthropicAI finally came back with a version that's above Opus and serves as their new flagship.

Opus 5.5 is still phenomenal, don't get me wrong, but it never really had that same level of understanding with nuance and ambiguity that Fable did. Think of a brilliant professor who's insanely good at whatever they do. It doesn't exactly mean they're good at communicating or teaching or otherwise; they're purely good at their specialty. That was Opus 5.5, which I've often found misunderstanding some of my intent and frequently exiting the scope that I set for it. The results were still excellent, but it wasn't ideal for what I wanted. That being said, being the most capable model, Opus 5.5 was still the best choice. Fable 5.2 however has both the brilliance of a genius professor, except to an even greater extent, AND it has a "big-model" level of understanding that Opus 5.5 just can't match.

I think using Fable 5.2 as an orchestrator for Opus 5.5 agents may be the way to go from this point onward (until a new release, that is).

[doghut @doghutty]

NEVER trust the AI "insiders". They couldn't even get Fable 5.2's number right smh.

[doghut @doghutty]

dayum dario and his gang r hella locked in rn, @OpenAI @thsottiaux @sama where u guys at??

[doghut @doghutty]

dude speak for yourself, having an affair is NOT a “mistake any of us could make” 💀. don’t get me wrong, i like the guy, but this was a REALLY bad mistake, and naturally there’ll be consequences that he’ll have to contend with both publicly and privately. it’s not just any other “we’re all human” mistake.

[doghut @doghutty]

@SingaporeAir was (and still remains) one of my favorite airlines in the world, but I'll have to agree with a lot of the critics on the new business class seat. Originally, I assumed people were overreacting, until I checked out The Points Guy's video showcase, and uh, yeah it actually is really muted.

I'm just not sure why they steered away from SQ's theme so much. It feels really cold. And they did a phenomenal job with their first class product, so I'm not sure what went so wrong with business, economy, and premium economy.

If anything, I'm pretty sure the new business class is a regression from the original business class product, which although a bit dated, is highly functional, well designed (especially with how the bedding is able to be stored "behind" the seat from the smart folding design), and spacious. The new business class appears significantly more crammed.

I'm not exactly against newer/more modern minimalistic/muted colors either. For instance, I think British Airways' Club Suites and Starlux's business class are perfectly fine, despite darker tones. SQ's new business class here just doesn't sit right.

[doghut @doghutty]

In all fairness, I think a lot of this sentiment also comes from (1) Anthropic and OpenAI themselves perpetuating a lot of the fear mongering and AI dooming, so they practically were asking for this, and (2) the AI slop problem, which has thoroughly invaded and saturated the internet at this point. Products like Claude Code and Codex are especially appealing to us people who like to build, which is why a good chunk of people in engineering and CS and other tech/STEM related fields are perfectly aware of their existence and arguably account for a significant share of usage as well, but most of the population isn’t like this. I doubt most people in the US (and the rest of the world) even know the basics of VS Code or a CLI or anything of that sort. They just don’t care, so AI in their eyes is a net negative.

[doghut @doghutty]

I don't get why guys like @kimmonismus and @TokenGremlin (nothing against them, just an observation) keep over-reading things on this platform. Of course 6.2 Astra wasn't going to release today. There were barely any indications to begin with.

@thsottiaux's vagueposts are just regular vagueposts. People forget that it's still his own account, and he gets to post whatever geeky or quirky or autistic thing he wants to post. These AI guys on here go "OMG HE SAID 'LE GPT FAT' OR SMTH HE MUST BE HINTING AT THE RELEASE OF 6.2 ASTRA." Let the guy be. Not all his posts need to represent OpenAI.

[doghut @doghutty]

they're pacing the frontier a bit too much 😔

[doghut @doghutty]

@TokenGremlin or openai’s just offering more generous packages to get more talent

[doghut @doghutty]

@TokenGremlin or… hear me out… people switching roles at different companies and trying new things is an entirely normal and common phenomenon, and people are reading far too deeply into it 🤷‍♂️ just my two cents