They’re toys with some specific productivity applications that are almost entirely unrelated to the chatbot elements. That’s it.

A lot of people have been fooled into thinking these are good for looking up information because they don’t remember a time when internet search worked. They don’t remember the days when you could use the right keywords and synthax and other tricks to get the correct search results, they started using the search engines after SEO and enshitification made it almost unusable.

regardless of if you are right/wrong/too-early on this conclusion, the mass enshitification shall continue. sure, bandwidth costs due to AI is annoying, but the real reason for ID checks and login walls is to prevent bring-your-own (API key or self-hosted LLM) apps from working, because working BYOK means not paying for their $20/month LLM subscription AND that would mean they are not selling your usage data AND it means they are not selling your ID verification data to Palantir.

we are eternally 2 years away from agi

That’s like how we are eternally 30 years away from nuclear fusion.

I got a new job in the last year and my boss is … friendly. He owns this small business, so he is what he is. There are a handful of people at this company with decent programming skills. It is not a software company, but has a few internal tools and small web applications. I am one of the programmers. Lately he has gotten the AI bug quite badly. He thinks the old way of software is just going to go away and AI just going to do everything in the future. Honestly, I have used Claude Code to vibe some stuff for work that I hate doing and I can get pretty decent results, but I am sort of operating as a senior developer giving assignments to a junior, reviewing its decisions, giving feedback, giving goals and requirements for maintainability, making sure each iteration is on track, stuff like that. I have a lot of experience so I know what the results should look like. But I also like to write good quality code, so I still write a lot myself.

So because the boss/owner thinks Code is Solved with AI, we’re giving a bunch of people at the company with no programming skills and frankly average at best basic computing skills Claude Code and asking them to try having it write tools to automate parts of their workflow. IT is horrified. I am morbidly curious. Like I said, I can get it to do things because I am guiding it at every step. The boss is simply not going to understand that the shit they’re generating is unmaintainable trash and that code that nobody can understand is a liability. He once showed me something he had it write and it was like a browser mockup of a user interface that obviously did nothing pertaining to the task it needed to do.

He thinks basic software engineering fundamentals are The Old Way. But computer science is real. Code generated by AI is still subject to the same principles as human generated code. No matter how good it gets at generating code, it will still need structure. The potential meanings of a certain quantity of lines of unstructured slop is still intractable for any computer humans could ever potentially build. Code needs to be verifiable in some way and made of components that can be understood individually. Algorithms have provable properties. Not to mention that the end result of success here is a company with no institutional knowledge of the business or skills. There is no reason for such a company to exist. Anyway, I love being essentially told that the entire field of computer science is The Old Way.

I’m very AI-skeptical in terms of the incredibly negative societal impacts of it, but as least in the field that I most associate myself with (software engineering, but I’ve been unemployed for more than 2 years), they’re somewhere between very useful and wildly useful. Most of my friends who are employed in this field haven’t written a single like of code by hand in months, and spend at least $1000/month on behalf of their company on tokens (per engineer). It’s the single biggest change to the field that has happened in at least 20 years.

I think that a big part of the crazy valuations come from the idea that this will be able to replace lots of labor, as least in the tech sector, if not more broadly in other blue collar sectors. In terms of software engineering, at least, I don’t think it works to replace engineers wholesale, but I do think that it can work as a productivity multiplier, where now one engineer can do the work of what used to take five to ten.

(I think the valuations of the AI companies are still completely ridiculous, because they all rely on the idea that one of these companies, i.e. Anthropic or OpenAI, will achieve a monopoly on having the useful model, which, outside of AGI , will never happen. Firstly, I think that currently they are both top tier models, so one would have to extinguish the other, and secondly, the Chinese AI labs are pretty close behind, or in some cases even better, and they release their weights so there’s no lock-in.)

Outside of tech, I’m a lot less sure. Right now, writing software is the main thing they’re being trained on, and they’re getting better and better with each release. Once they hit a wall with that, these companies might focus their training efforts on other tasks, and it would likely get better at those too. I’m not sure what areas would come next (I’m sure it would be based off of a combination of what they think will make them the most money and what is easiest to get large amounts of training data for), but it has the potential to do a lot of white collar computer jobs in the near-ish future imo.

I agree with a lot of other points in the thread, but I think they do correspond to a pretty substantial shift in the field of tech, which is a big part of the US’s economy.

I literally haven’t used an LLM in 5 years, only as a novelty “look at this funny thing that can’t even form a sentence right”. But I have to live with AI-generated code at work and holy shit, I find I can’t communicate to the people that put it there. Like on a vocabulary level, it’s that bad. The worst part is I don’t know if a different job would be even worse, and so I’m in a purgatory where I can’t say I’m doing badly but I’m unhappy and, more generally in life, stuck.

I guess I’m paying for the “don’t be the best person in the room” part or however it goes. I can’t do anything exciting because I’m too busy fixing fuckups

I do think there will be a fundamental change in the way people interact with computers, now that AI is being shoved in lieu of an UI. For many non techie people I know, being able to talk to the computer using natural language, and more or less get the result they want is a huge improvement over learning the steps to follow to get the same result by rote memorization (because unlike me, they don’t care about learning why the computer follows the steps in that way, and that’s fine)

The best case scenario for chatbots and LLMs in my opinion is for them to be transformational as an interface, rather than as a key part in applications’ inner workings. I see them much more in the same way as smartphones, the mouse, or graphical interfaces changed the way people interact with tech, and apps had to change to fit this new interface paradigm.

Of course, a mouse doesn’t have a 25% chance of moving to somewhere you don’t want it to or to click randomly. And they don’t take the equivalent of a country’s worth of energy to produce and maintain, either.

Kimi has been consistently useful for general purpose inquiries/investigations and handled everything I’ve prompted without mistakes.

You can avoid the reddit speak by giving saving a directive in your settings (kimi calls these memories) specifying the tone and style you’d prefer it use in its outputs. I instructed it to speak to me like the ship’s computer in star trek TNG, basically. No fluff.

As for accuracy, the “thinking effort -> high” setting is default and causes it to search the internet for references before answering the query rather than using just what’s baked into its model. I assume that’s a big factor it’s accuracy. I’ve even seen it automatically write and execute python scripts to aid in answering a prompt.

Scenario where they do achieve a grand hyper intelligence but it’s so owned by the capitalist class that asking it to solve issues such as climate change leads it to saying that as an AI model it’s not allowed to discuss sensitive political topics.

I still have to read the comments on this post, but I thought of making one. I mostly agree actually. I only used AI chatbots a handful of times, but it gets old really quickly for me, specially because it treats you like you’re incapable of making a mistake, the user has to be right and must be treated as absolute royalty. It’s not like it can’t be useful, it can, it has helped me in some situations, but others it just shat the bed.

I haven’t tested any of these western AIs tho, I only used DeepSeek and Kimi, so I can’t compare them to that, but still. With the exception of Gemini, because Google just forces it whenever you search something. This one has been useful in one single thing, aggregating a bunch of sources neatly in some cases so I don’t have to keep scrolling until I find what I want, but it also just misses spectacularly and invents things regularly. Also out of all of these, Kimi has been the one that’s consistently better than the others in my experience.

I can’t really differentiate among any of the models cause if one doesn’t work I’ll just toggle to a different model on duckduckgo. They tend to work very well for my purposes, which is locating sources of info that a regular search doesn’t turn up (or the results would take a lot of time to parse through).

Their shortcomings aren’t a big deal for this because if one returns an incorrect result or hallucinates I can just correct it or hop models. And the ability to do fuzzy searches using plain language descriptions engines have never been good at, so in that case the LLMs are straight up better. But yeah otherwise a lot of their role for me is filling in for how bad search engines have gotten.

I only code a little bit to solve my own problems so I’ve left it to people who’ve actually investigated to have an opinion on that angle. But last week I needed to add a quick function to one of my things so I tried one of the models, and it knocked it out immediately with no need for corrections. Less fun than solving the problem myself but woroed in a pinch since I was busy.

I need to look into local models again now that they’re getting good and lightweight. But I’m not sure I’d be able to get much utility since my sole use case right now is search and I’m running an old thinkpad. One of these days I’ll tinker with it.