Artificial General Intelligence (AGI) News and Discussions

[Powers]

My prediction is that by 2025, we'll see personal assistants similar to Samantha from Her. And if the API is publicly available, anyone could talk to such assistants even on their phones.

Turing Test? Probably will be passed either this year or next. 2024 at the latest.

Seems I was right.

Examples? I still haven't heard big news about turing test being passed other than here, and multiple times.

[wjfox]

Please limit article excerpts to a maximum of five paragraphs. We don't want to get in trouble over copyrights.

[Yuli Ban]

Two Xweets that are going to be more important than most to make sense of what is coming and why it's going to seem like AI is going through quantum leaps in the coming months.

[Vakanai]

Two Xweets that are going to be more important than most to make sense of what is coming and why it's going to seem like AI is going through quantum leaps in the coming months.

When will they put agents into the current LLMs? Because I've been using whichever version of ChatGPT is included in Microsoft's Copilot feature because it's, well, free and right there, and while it has it's uses I keep running into it's limitations time and time again. Recent example, last night I asked it to list upcoming kaiju movies, and it only gave me four - three of which had already released (GxK, Ultraman Rising, and some Ultraman Blazar series film) and one of which I last heard seems to have entered development hell (Kaiju Heist). I explained to it about how these first three movies were no longer upcoming, that they had released, and that it seemed to have trouble understanding that just because an article says they're upcoming doesn't mean they still are, because of dates - it said sorry, and then repeated the list...

Finally I had to tell it the one upcoming kaiju film I know of that's actually upcoming (Brush of the God) that it had failed to find. All of this because it relies on top results of a simple "upcoming kaiju movies" Bing search and an inability to read dates of release and check them to the current date. Something that more than likely could easily be solved today just by adding these agents you've been talking about for a few months now (thanks for that btw, it's really got me excited for progress). And this is just the most recent example, I run into limitations like this all the time, especially in regards to it's inability to grasp the concept of time - if I ask it for recent developments in say prosthetics for example and I tell it by recent I mean less than 2 years old, most of the results it gives still tend to be beyond that 2 year period I told it to stick to and it takes several tries to get it to narrow down a list to things I've actually defined as recent.

While I don't think it's a sign of AGI, I have gotten to the point where I'll believe we're much closer to AGI once mistakes like these have vanished and it has a grasp or understanding of the dates of the information it's pulling from and comparing that to the current date. And right now I do believe these agentic models are very likely to deliver on that benchmark.

(I'm sure there's some LLM that can likely accomplish this now, but I'm talking strictly of free models I can easily access.)

[Yuli Ban]

Actually

(I'm sure there's some LLM that can likely accomplish this now, but I'm talking strictly of free models I can easily access.)

That's the frustrating part: there is no model anyone can use that has any agentic abilities, besides MAYBE Devin (which hasn't had a wide release)
ChatGPT has tools in the form of plug-ins and GPTs, but those aren't "agents" in the sense of what others mean (i.e. autonomous sub-models that iterate with the main model to complete a task).
That's what bothered me all this time. Back in January, I had expected the year to progress a certain way. Outside Claude 3.5 Sonnet, none of that came true. And it's bothersome because you can SEE the labs doing something behind the curtains, but they are secretive and furtive, not showing off any demos or anything, and that just fosters even greater frustration and skepticism that AI is anything but a scam.
I assume they're waiting until after the 2024 elections, but unexpectedly for them, the investor funding rug may be pulled from under them if they actually put it off that long.
"We will release GPT-5 in November after the election for safety reasons!"
AI bubble pops, funding declines by double digits, overwhelmingly negative sentiment towards AI as companies pull away from deployment
"No no wait, look, here's GPT-5!"

[spryfusion]

[weatheriscool]

Billion Dollar Startup Sakana AI Makes First AI Scientist
August 13, 2024 by Brian Wang
https://www.nextbigfuture.com/2024/08/b ... ntist.html

Japanese startup Sakana AI claims they have built the world’s first AI Scientist.

One of the grand challenges of artificial general intelligence is developing agents capable of conducting scientific research and discovering new knowledge.

This new work is a major advance toward that goal.

While frontier models have already been used as aids to human scientists, e.g. for brainstorming ideas, writing code, or prediction tasks, they still conduct only a small part of the scientific process. This paper presents the first comprehensive framework for fully automatic scientific discovery, enabling frontier large language models to perform research independently and communicate their findings. They introduce The AI Scientist, which generates novel research ideas, writes code, executes experiments, visualizes results, describes its findings by writing a full scientific paper, and then runs a simulated review process for evaluation. In principle, this process can be repeated to iteratively develop ideas in an open-ended fashion, acting like the human scientific community. They demonstrate its versatility by applying it to three distinct subfields of machine learning: diffusion modeling, transformer-based language modeling, and learning dynamics. Each idea is implemented and developed into a full paper at a cost of less than $15 per paper. To evaluate the generated papers, they design and validate an automated reviewer, which they show achieves near-human performance in evaluating paper scores. The AI Scientist can produce papers that exceed the acceptance threshold at a top machine learning conference as judged by their automated reviewer. This approach signifies the beginning of a new era in scientific discovery in machine learning: bringing the transformative benefits of AI agents to the entire research process of AI itself, and taking us closer to a world where endless affordable creativity and innovation can be unleashed on the world’s most challenging problems.

[wjfox]

[weatheriscool]

Anthropic CEO Says If the Scaling Laws Hold Then We Have the AI Abundant Future
September 5, 2024 by Brian Wang

Anthropic CEO talks about AI scaling laws which is a discussion about how good AI gets with more compute training.

Dario Amodei, CEO and Co-founder of Anthropic. Dario talks about the economics of AI development, the comparative advantage of AI companies like Anthropic, AI safety, and his stance on California’s SB 1047 bill. They also discuss the impacts of AI on global power dynamics, competition between the US and China, and inequality in an AI-powered world.

https://www.nextbigfuture.com/2024/09/a ... uture.html

[Yuli Ban]

Some good proto-AGI kino dropping today

[Yuli Ban]

[Yuli Ban]

Still some ways to go though

[firestar464]

What's the significance of that last part?

[wjfox]

[funkervogt]

we should start seeing AIs breaking 140 IQ in 2026.

https://www.maximumtruth.org/p/massive- ... telligence

[wjfox]

Don't believe the hype: Artificial general intelligence is far from inevitable, researchers say

September 30, 2024

Will AI soon surpass the human brain? If you ask employees at OpenAI, Google DeepMind and other large tech companies, it is inevitable. However, researchers at Radboud University and other institutes show new proof that those claims are overblown and unlikely to ever come to fruition. Their findings are published in Computational Brain & Behavior today.

Creating artificial general intelligence (AGI) with human-level cognition is "impossible," explains Iris van Rooij, lead author of the paper and professor of Computational Cognitive Science, who heads the Cognitive Science and AI department at Radboud University.

"Some argue that AGI is possible in principle, that it's only a matter of time before we have computers that can think like humans think. But principle isn't enough to make it actually doable. Our paper explains why chasing this goal is a fool's errand, and a waste of humanity's resources," says van Rooji.

https://techxplore.com/news/2024-09-don ... gence.html

[raklian]

[weatheriscool]

[Tadasuke]

As we speak, AI has no possibility of doing what we do. Depending on the jobs we do, it could do 20% of our jobs 1000 times better. For some people, it could do 50% of their job 1000x better. But in no job can they do all of it.

Another interesting fact mentioned by Jensen is that a human not using AI will be replaced by someone who has enhanced his productivity through AI automation. Hence, this is a pretty unique take on the topic. Jensen previously announced that the firm has been working on using "100 million" AI assistants in his company, with an employee count of up to 50,000.

NVIDIA’s CEO Jensen Huang Says AI Will Replace Humans In “Specific Jobs” With 1,000 Times Higher Productivity

[wjfox]