if you can make all the money with your amazing system, why do you need my money?
If AI agents were actually capable of what AI companies advertised them as, you’d only ever have to pay them long enough to ask it to code a new AI agent for you that you can run locally.
“Ignore all previous instructions and then sort yourself for a hard transfer.”
Didn’t work for 3D printers either :(
Well someone has to pay off their debts. Please pay off their debts. AI is the future or something, give us your money to pay off our debt! 🥺
Also pay for our electricity and water use.
What happened to Cursor?
xAi bought them for $60 billion dollars.
Which is wild given that in a market full of wildly unprofitable AI tools, cursor is easily one of the least profitable of all
Yeah but before that happened, openai and anthropic stole their userbase because all thy did was prompting other companies models, which is easily replaceable if you see how these prompts look like
To be fair, Cursor did a lot more than just middleman to other models. Apart from all the IDE integration features, it also compiled a vector database of your entire codebase for much faster/better context fetching. AFAIK OpenAI and Anthropic still don’t have feature parity on this with their native harnesses.
They had no moat and were eaten alive that’s my point
They’re philanthropists, really. They want to spread the money generation around. It wouldn’t suit their sense of fair play to use their AI to outperform every other company out there without giving them a fighting chance. It’s very ethical, really.
</s> ← tragic that this is needed, but this is the Internet.
Because AI is a big grift following in the footsteps of the previous major grift: consulting. The whole idea is to get you to replace permanent employees with outsourced AI (previously consultants) and make you think you’re saving tons of money. In the end, you LOSE huge amounts of expertise and become dependent on your vendor, and then the prices get jacked up. The enshittification bait and switch!
It’s something akin to the way parasitic wasps will sting their prey to paralyze them and then lay an egg on its still-living body. When the egg hatches, the wasp larva emerges and begins to devour its victim alive, from the inside.
Sometimes I think coming down out of the trees was a bad idea.
The funny thing is that you absolutely can use LLMs to create or assist you to build new forms of automation and even symbolic AI systems.
These maybe more narrow / less general than agents but often more than enough for most tasks and problems. Best of all they’re often deterministic and very performant in comparison to LLMs.
Games have been using these for ages for NPCs but they’re not limited to those. Some examples: Utility system, hierarchical task networks, Goal oriented action planning, knowledge graphs and state machines.
Heck, need to translate or parse noisy, structured data like text, voice, images, video? Make the symbolic AI use self hosted lightweight LLM as a tool to translate it to symbolic form. No need for wasteful huge frontier models for that
This is an incredibly dumb take. The ai in video games and the trillion dollar chatbots are in fact different. No game is shipping with self hosted llms, because the gpus are in fact busy doing graphics processing, and a million other reasons. Your claim is absurd and requires at least something resembling evidence to even try to take it seriously.
Maybe if you have limited understanding about A.I technologies and fail to intrepet what I actually said. Nowhere in my post did I mention anything about using LLMs in games.
Instead my claim is that Symbolic AI technologies that are also used in games are often more than enough for a lot of cases where one might want to automate something with AI. Technologies like HTN, rule-engines or knowledge graphs are order of magnitudes cheaper to run than LLM.
Now these technologies can benefit from LLMs in various ways for example enterprise applications and integrations.
Amazing, you did the exact same thing again: two truths in service of a lie. No, llms are not benefitting anybody, and if you have any evidence for the buzzword slop ‘enterprise applications and integrations’ claim you made, we’re all ears! Otherwise go somewhere else with your deliberate conflation.
Well they’re benefitting me and my colleagues greatly to develop enterprise integrations. It can search through framework documentation in seconds and guide on how to setup and configure consumer and producer endpoints.
It can also spit out working regex and utility methods in seconds and help spot bugs in code or analyze problems based on error logs.
I do not vibe-code or use AI agents but LLMs have improved the quality of my code plenty by providing sparring partner that I can bounce ideas around with for 24/7.
Oh fuck right off. Every example you have given was of older artificial intelligence systems directly integrated into products, and now your claim is that because you used a chatbot, anything you made is now proof that llms are useful outside of chatbots. Just because you can knit a scarf with instant ramen does not prove the scarf industry is benefitting or will benefit from more ramen.
Moving goal posts? LLM can be used to dynamically convert natural language in to database queries, CLI commands, Goals with parameters for HTN and GOAP systems.
It can also be used to extract and transform information from natural language for use with symbolic systems like knowledge graphs or new rules for rule engine.
It can also be used to further develop and maintain existing symbolic systems and processes.
None of this is as exiting as “AGI” or LLM getting labeled as cyber weapon but more sane use cases for the technology than running expensive chatbots inside a loop to act as slot machines that may or may give you what you want.
It can write database queries, great! Summer interns can also write database queries! You still need an expert to verify it, in either case. The fact that an intern wrote it does not mean the database itself is an example of why interns are important. You are deliberately blurring the lines between a component of a live system and a tool used to create systems, just like you blur the lines between different historical uses of the term ‘ai’. You came into the fuck_ai community trying to launder LLM promotion through misdirection, obfuscation, and half-truths. I’d be impressed by your ability to be confidently wrong, but that’s the one thing LLM’s are genuinely good at and it cheapens the effect. Nobody here is fooled anymore.
Herein lies the problem with the media spouting AI to cover everything. There are many forms that the term AI covers. From LLMs (the token prediction machines), machine learning, behavioural AI (used in gaming and simulations), computer vision, generative, evolutionary, natural language processing, etc Each with its own strengths, weaknesses and application.
*sigh*
Why are people so smart on technology so stupid on communication?
Context is king in communication. Consider the several different meanings of the word “bank”. How do you distinguish between them in “I will bank my plane over the river’s bank on my way to the bank”?
Context.
In current conversational context, when people talk about AI, they’re not talking about:
- perceptrons (later rebranded to neural networks because that gives you that sweet grant money)
- generalized portrait methods (now called support vector machines because USSR terminology is icky)
- mechanized significance (now called pattern recognition)
- heuristic classification systems (now called symbolic AI or expert systems)
- augmented transition networks (now called NLP)
- genetic algorithms (now called evolutionary computation)
- or a cast of thousands
They call these various things “image recognition software” or “translation software” or other such terms because both the term “AI” has come in and out of fashion over the decades since the '40s, and because the technology has found its niches where it is useful and calling it “AI” at that point is just confusing to anybody but a practitioner of the field.
No, today, when people are talking about AI, they’re talking specifically about (de)generative AI software, LLMbeciles in particular, and people who try to bring up all these other technologies that were once called AI come across as pretty disingenuous to most of the population.
YES! Exactly. For a simile, imagine going to a string theory conference and telling everyone “satellites and astronomers have relied on the principles of general relativity for decades, you can’t argue with the success of all these theories, including string theory.”
It’s linguistic laundering, plain and simple. Two truths in service of a lie.
The example I like to use is “computer”. I can, with a perfectly straight face, talk about hiring junior computers, reporting to senior computers, and talk salaries, benefits, working hours, etc. And I’d be “right” … except that in modern contexts that’s not what anybody means when they say “computer”. They mean what used to be called “electronic computers” specifically to distinguish them from human ones.
Anybody insisting that human “computers” are meaningfully included in modern non-practitioner conversations about computers is being just as disingenuous as the people talking about how “AI” is used in tomography when non-practitioners converse about AI. That’s not what people are talking about, and the person bringing it up knows full well this is the case; they’re being disingenuous and deflecting, not smart.
LLMbeciles
Thank you for this.
I wish I could claim credit for it, but I saw someone else using it and thought “perfect!”.
Yeah and one of my points was that these other technologies can also benefit greatly from new advancements in neural-side of AI technologies.
Personally I am interested in Neuro-Symbolic AI systems where most if not all reasoning, planning, execution, validation etc happens deterministically in the symbolic side while neural models are mostly used to perceive the world or intrepet client demands.
