I don't know if the investments in AI are worth it but am I blind for not seeing any hope for AGI any time soon.
Agentic AI is interesting perhaps but I hardly have had it work perfectly, I have to hold it's hand at everything.
People making random claims about AGI soon is really weakening my confidence in AI in general. Given I haven't seen much improvements in last few years other than better tools and wrappers, and models that work better with these tools and wrappers.
Anyway, here's something I've recently build that shows the HN consensus when it comes to AI-Coding (spoiler: they say it's quite good): https://is-ai-good-yet.com/ Is AI “good” yet? – A survey website that analyzes Hacker News sentiment toward AI coding.
I'll believe AI is approaching human equivalence when one writes the next great American novel, or becomes superstitious - if you're going to be human level, genuine creativity ex nihilo and/or pattern recognition that overfits into religious belief would convince me.
Source? From my knowledge, we do not "think in language", but we learn to finetune our thinking to be expressed in the form of words. Unless you consider pictures as language, after all "A picture is worth a thousand words"..
I mean. “Yes”, but no, not really.
I think one day this kind of system will probably exist in some form, I don't see any fundamental reason why not but to believe fraudsters like Amodei and Altman telling you that in 6 months world will possibly end and the AIs will take over, it's just nonsense fed to boomer investors who have 0 clue.
Many crypto scammers have pivoted to AI because there are new useful idiots to scam now that crypto is not doing too hot.
Frankly the whole world is just scam after scam nowadays, they noticed that the Elon Musk strategy of continuously promising something "the next year" actually works so they are all doing it stringing people along.
You could be wrong, hopefully. I'll just remain optimistic.
It's good that critical thinking still exists to keep us all grounded.
Edit: I think you meant to replace AGI with ASI.
What we have now is approaching the ability to solve problems generally with considerable effort but super intelligence is definitely out of reach for the time being.
Anyone still saying they'll reach AGI is pumping a stock price.
Separately and unrelated, companies and researchers are still attempting to reach AGI by replacing or augmenting LLMs with other modes of machine learning.
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...
It shows a remarkably consistent curve for AI completing increasingly difficult coding tasks over time. In fact, the curve is exponential, where the X axis is time and the Y axis is task difficulty as measured by how long a human would take to perform the task. The current value for 80% success rate is only 45 minutes, but if it continues to follow the exponential curve, it will only take 3 years and change to get to a full 40 hour human work week's worth of work. The 50% success rate graph is also interesting, as it's similarly exponential and is currently at 6 hours.
Of course, progress could fall off as LLMs hit various scaling limits or as the nature of the difficulty changes. But I for one predicted that progress would fall off before, and was wrong. (And there is nothing saying that progress can't speed up.)
On the other hand, I do find it a little suspicious that so many eggs are in the one basket of METR, prediction-wise.
I'm not convinced that "long" is equivalent to "difficult". Traditional computer can also solve tasks that would take extemely long for humans, but that doesn't make them intelligent.
This is not to say that this is useless, quite the opposite! Traditional computers shown that being able to shorten the time needed for certain tasks is extremely valuable, and AI shown this can be extender to other (but not necessarily all) tasks as well.
1. Ability to take large amounts of information into consideration, specifically large codebases (longer tasks usually involve larger codebases). LLMs struggle with this due to context window limitations.
2. Ability to make and execute on long-term plans. Also related to context window limitations, as well as what for a human would be called "executive functioning skills".
3. Consistency. If you have an x% chance to get stuck on each step of a multi-step task, then the more steps, the higher the failure rate. This is true for both LLMs and humans, but LLMs tend to have more random failures, both due to hallucinations and due to being worse at recovering if their initial attempt fails (they can have a hard time remembering what they're supposed to do differently).
These difficulties seem to generalize beyond coding to almost any kind of knowledge work. A system that could solve them all would be, if not AGI, at least a heck of a lot closer.
> I don't know if the investments in AI are worth it but am I blind for not seeing any hope for AGI any time soon.
> People making random claims about AGI soon is really weakening my confidence in AI in general.
The "people" that are screaming the loudest and making claims about AGI are the ones that have already invested lots of money into hundreds of so-called AI companies and then create false promises about AGI timelines.
Deepmind was the one that took AGI seriously first which it actually meant something until it became meaningless, when every single AI company after OpenAI raised billions in funding rounds over it.
No one can agree as to what "AGI" really means, It varies depending who you ask. But if you look at the actions made by these companies invested in AI, you can figure out what the true definition converges to, with some hints [0].
But it is completely different to what you think it is, and what they say it is.
People talk about AGI because this is how corporate marketing will create AGI, even if AGI is not near to what we could call possible.
But this is how things work now, corporate marketing says what is real and what is not.
What's your definition of AI, and why is yours righter than anyone else's?
All those media bring more or less the same concept of AI that is completely different from LLMs.
Corporate marketing transformed LLMs into AI telling people that what people think about AI is the same as LLMs when actually it's not.
Part of the problem with LLMs in particular is ambiguity -- this is poisonous to a language model. And English in particular is full of it. So another potential that is being explored is translating everything (with proper nuance) to another language that is more precise, or by rewriting training data to eliminate any ambiguities by using more exact English.
So there are ideas and people are still at it. After all, it usually takes decades to fully exploit any new technology. I don't expect that to be any different with models.
Right now it seems like this could lead to AGI very fast (5-15 years).
Especially because the richest companies in the world are investing extrem amounts of money and are very fast in pivoting into it.
Image and video gen alone made crazy progress.
We have capabilities today which were unthinkable just a few years back
Is it perfect? No. Far from it. Is it useful in some, and in the future many situations, Yes.
Having said that, I could not care less about AGI and don't see how it's any relevant to what I wanna do with AI.
Context rot is real and people who complain about AI's hallucinating and running random wild, I don't see it when the context window is managed properly.
> If builders built buildings the way programmers wrote programs, then the first woodpecker that came along would destroy civilization.
I think you’re taking my analogy too literally. I just mean they help you go faster. When building software you have a huge advantage in that there is very little risk in exploring an idea. You can’t hurt yourself in the process. You can’t waste materials. You can always instantly go back to a previous state. You can explore multiple options simultaneously with a favorable cost to doing so.
You don’t have to let your standards drop. Just consider AI coding an interactive act of refinement. Keep coding manually where you meet too much resistance. Accept the LLM can only do so much and you can’t often predict when or why it will fail. Review its output. Rewrite it if you like.
Everything always has a chance of being wrong whether or not you use AI. Understand an AI getting something wrong with your code because of statistical noise is not user error. It’s not a complete failure of the system either.
It’s a mega-library that inlines ether an adjustment of a common bit of code or makes up something it thinks looks good. The game is in finding a situation and set of rules which provide a favorable return on the time you put into it.
Imagine if LLMs were right 99% of the time, magically doing most tasks of a certain complexity 10x faster than you could do them. Even when it’s wrong you will only waste so much time fixing 1% of the AI’s work. So it’s a net positive. Find a system that works for you and lets you find something where it makes sense to use it. Maybe 50% of the time and 3X faster than you makes it make sense.
In some domains you can absolutely learn some basic rules for AI use that make it a net positive right away. Like as a boilerplate writer, or code-code translator. You can find other high success likelihood tasks and add them to the list of things you’ll use AI for. These categories can be narrow or wide.
I think a lot of confusion with skeptics is they think - oh someone's invented the LLM algorithm but it's not that good - what's the big deal?
The people who think it's coming eg. Musk, Altman, Kurzweil, the Wait Not Why guy and myself tend to think of it more coming down to hardware - the brain's a biological computer and as computer hardware get's faster each year it'll overtake at some stage. The current backprop algo was invented around 1982. It works now because hardware.
Also the present algorithms are a bit lacking but now we have the hardware to run better algorithms, billions of dollars and many of the best minds are getting thrown at that. Before the hardware was there there wasn't much financial motivation to do so. So I think things will advance quite quickly there.
(Wait But Why thing from eleven years ago. Has cartoons. Predicted human level about 2025 https://waitbutwhy.com/2015/01/artificial-intelligence-revol...
Moravec's 1997 paper "When will computer hardware match the human brain?" Quite science based - predicted "required hardware will be available in cheap machines in the 2020s" https://jetpress.org/volume1/moravec.pdf
And here we are.)
Crazy how fast people acclimate to sci-fi tech.
What we have today in the form of LLMs would be a VI under Mass Effect's rules, and not a very good one.
I agree that the term AGI is the problem. If I have something as intelligent as a mouse that should be AGI, if I have something a intelligent as a bird that should be AGI. Same is it's as intelligent as a 2 year old human, or someone with an IQ of 75. Those would all clearly be Artificial, General Intelligences.
But the problem is the term AGI also oddly has this bar that if must be equal to our better than human (a standard that the majority of humans would fail based on definition of intelligence alone). Plus multidisciplinary better than all humans (which it would take a super genius to have a human accomplish).
Given the current definition of you took a bright high schooler and made them artificial they wouldn't count as AGI which makes the definition silly.
And that is separate from the entire concept of sentience - which it's unclear if it's a requirement for intelligence.
It's all a bunch of squishy definitions mashed together.
idk ... even sam altman talked a lot about AGI *) recently ...
*) ads generated income
*bruhahaha* ... ;^)
just my 0.02€
See it as a simple BS detector.
Then, in March 2023, with GPT-4, I said that we'll have AGI only ten years later, and the progress in the last few years (multimodal stuff, reasoning, coding agents) hasn't changed this view.
We aren’t even really in “minimum wage job” territory yet, never mind a median salaried employee.
I’m still being paid a small fortune even though AGI is supposedly available for the cost of a monthly Starbucks habit.
I recently had to talk to a real human to get a cell phone activation/payment to go through even though supposedly an AI should be better at communicating with digital payment systems than a human clicking around with a mouse. The only purpose the AI in the system had was to regurgitate support articles and discourage me from contacting the human who could solve my problem.
As far as I can tell, HN defines an AGI as something that can do all the things a human can do better than a human. Or to put it another way if there is something the AGI can't do better than a human expert, then it will be loudly pointed to as evidence we haven't developed a true AGI yet.
Meanwhile I'm pretty sure the AI firms are using a very simple definition of AGI to justify their stock price: an AGI is an AI that can create other AI's faster / more cheaply than their own engineers can. Once that barrier is broken you task the AGI with building a better version it itself. Rinse, lather and repeat a few times, and they dominate the market with the best AI's. Repeat many more times and the universe becomes paperclips.
