It’s a bit questionable, IMO, where we’re going with AI. I mean we want a problem-solving tool. But that’s not really what we have. We have a text generator instead. With a lot of limitations and really weird quirks. Now we add a lot of duct-tape on top. Try to minimize stuff in the harness/framework, add guardrails, citations, make another text generator fact check the first one… But that’s all still just something that isn’t a proper problem-solving machine at the core, with lots of hot glue and duct tape around. And once it’s used for more complex tasks, it all shows, and issues always add up. Plus the framework is notoriously bad at fixing edge-cases and stuff like that. So it’s very limited in what it can do and doesn’t have good accuracy.

I think the only proper fix is to dig down and really fix the issues. Come up with a way to not make it make up wrong facts. But that remains to be an unsolved challenge.

Does this happen to people? This article sounds a bit like it’s based on the experience I had with satnav 15 years ago. These days it’s absurdly precise and the “educated guesses” are correct almost all the time. With one issue remaining: Once I specifically don’t follow directions and end up on the wrong parallel road, or ramp, it takes a while to disentangle the educated guess and gps position and make up it’s mind where I’m at. But that’s pretty much it. And it’ll manage to cope after a few seconds.

Just to let you know: There’s some maximum comment depth here. Our conversation might be cut short, soon.

Idk man. Just to let you know: If you ever feel like you want to change your ways… repent… or just talk to someone… learn about world as it’s out there in reality… Just ask.

Feels to me like you’re deeply unsure about your own identity, so you’re afraid of other people’s ways, their gender, sexuality, religion… You feel threatened and miserable… But it doesn’t have to be that way! You could also be proud of who you are yourself. Focus on your own happiness, relationships, religion, or absence thereof. This is the way. And it’s working out great, where I’m at. Also as a bonus, it makes you feel a bit less depressed about the world. You can do it as well! Takes a bit of work, though, it doesn’t come entirely for free.

I told you half an hour ago: Central Europe.

But you get me super curious. What do they teach people on the moon, where you’re at, about Europe?

Totally not a gay place? Just ignore all the rainbows painted on the pavement in all major cities? Also ignore all the men’s sauna clubs, night clubs where people of all genders are having gay sex in the basement? Nude beaches? Their stupid worker’s unions and healthcare system which all totally isn’t gay at all? Ancient Greeks and french cuisine, not remotely gay? It’s just the “tiny slice of the country” @hendrik’s at? Everything else is totally serious and straight?

Have you ever been here? Or do you just pull your world view out your arse?

No it’s not. We’ve come a long way since whatever time in history you’re stuck at.

We’ve been proud an loud for quite a while now. And everyone but you is enjoying being gay and a furry and gender-fluid and vegan and everything. Big party here. I’m a bit sad you don’t wanna come…

Cute, how someone still thinks “gay” is a derogatory term and the other person must feel insulted… You should have talked to me in the 90s when I was like 11yo… Now I’m not. But I’m definitely a sweet transvestite. …from transsexual … Transylvania… 🎵🎶🎶 Ahahaaa 🎵

What was your post about, then? Other than you’re in the wrong place here?

Sure, why not? You got lost. You’ve been looking for some place. And you ended up in another place which wasn’t what you were looking for. Now your strategy is to roll down your window and yell at the natives… 🤔🤔🤔

Lol. What’s up with you? You f-ed up?! That’s what your post is about?!?? I’m not you??

They’re doing it the right way, though! Social media is made for speaking to other people. I’d argue that’s the main point about it. Consuming news, less so… I mean people also use it for that, but I think you’d be better off using a RSS reader. At least that’s what I’m doing. You’ll even get the news the moment they’re published and from your favorite IT magazine / Android blog/newssite.

2027 Year of the Linux desktop…

This creates perverse incentives for sycophancy to persist: The very feature that causes harm also drives engagement.

I bet that’s also why AI is like that (sycophantic). I don’t see any technical reason why a language predictor needs to write overly agreeable text. Or be as bland and repetitive as they are. That’s probably because they’re designed (by big tech) to be like that. Due to user preference.

Idk, I used to be fascinated by language models early on, for example when Meta’s first LLaMA model weights got leaked. And the time after when we got new models and discoveries every other day. And if I remember correctly, they weren’t all like that?! I distinctively remember a few language models which were more or less agreeable than others. Some would respond to a question in a single paragraph. While ChatGPT loves to write pages of redundant stuff and claim to highlight every side of each coin.

But with each new iteration, all the language models homed in more and more on the “helpful assistant” style. Possibly pioneered by ChatGPT?! At least that one felt from the get go like it was supposed to treat me as a 5yo who needs a lot of fake positive reinforcement.

Interesting. I wouldn’t have expected commercial AI assistants to compute a lot of stuff on device. For speed / response time reasons, and battery drain. Other than maybe speech recognition… Guess I’m not up to date anymore. I haven’t tried Gemini. Most fancy stuff I use is speech recognition and Google Lens, respectively “Offline Translator” to translate Chinese, Japanese or Korean writing on products in the Asia supermarket. That app is perfectly alright with just the 6GB on my phone. But yeah, generative AI in form of a large language model or an image generator won’t fit. I’ve tried a few other open-source local AI apps and it can’t do any of the usable AI models.

Does anyone realistically use 16GB of memory on their phone? I think my Pixel 8a has 6GB and I think I’m alright?

Yeah, as I said, they love licenses. Their code is FUTO Source First License 1.kb(?). Their models FUTO Model Weights License 1.0. The dataset MIT and seems the training code is proprietary. That’s 4 differently licensed parts that go into one product.

Thanks, I forgot about that 😟

BTW, there’s other voice input apps. I changed phones since I last tried but there were other good ones which were more responsive than FUTO’s software. My F-Droid shows Sayboard, Outspoke (via IzzyOnDroid’s Repo) and Whisper+

Sure. I just installed the Keyboard. I’ll try it for a few days. But yeah. Usually I try to stick to Free Software, so there’s a good chance I’m gonna remove it from my device again. Unless they manage too get me hooked on swipe typing 😅 But I think Heliboard also uses a closed-source library for swipe typing.

Yeah, it’a been a while, but last time I tried swipe typing with Heliboard, it wasn’t good. I forgot what issue was, but I uninstalled it after a few days.

Interesting indeed. Their dataset is under MIT license. Their resulting models under a special license.

And I have no clue if there’s code available to customize it, train or fine-tune the models. And what license that would entail.

They often do “interesting” stuff with licensing. Pretty much since the beginning of FUTO. I’d also question the choice of wording in the MIT license. Swipe patterns aren’t exactly “software”. And I wonder if their community knew they were releasing their contributions under some license. I can’t find any paragraph detailing it on the current data contribution page.

Just a bit of a shame how they always frame it in their public relations. Of course they talk about the MIT license in their videos. They seem to known that’s good PR. But they casually drop all the rest and don’t talk about the license of what they did. Or if it’s even actually possible to adapt it to other projects, given that requires some additional code and not just the dataset or weights.