Artificial General Intelligence (AGI) News and Discussions

[firestar464]

Ain't that ASI by many definitions?

[weatheriscool]

Ghost in the machine: The race toward conscious, sentient AI
By Loz Blain
January 22, 2025

As science wrestles with the thorny job of proving, defining and understanding human consciousness, one pioneering startup aims to apply cutting-edge theories of consciousness to AI models – and create the first sentient AI. This is wild stuff.

The last time we spoke to Suzanne Gildert, she was, as it turned out, just days away from leaving her former post as Co-Founder and CTO at Sanctuary AI, where she oversaw the development of the extraordinary Phoenix humanoid robot, one of the most advanced AI-driven robots in an exploding market.

\
https://newatlas.com/ai-humanoids/ai-co ... c-gildert/

[firestar464]

ah yes what could go wrong

[spryfusion]

[firestar464]

AI versus the brain and the race for general intelligence

https://arstechnica.com/science/2025/03 ... elligence/

Excellent points here. Key take away is that an AGI needs to be able to learn on the job

[spryfusion]

[weatheriscool]

Bill Gates says AI will replace doctors, teachers within 10 years -- and claims humans won't be needed 'for most things'
https://share.newsbreak.com/cban9msq

[spryfusion]

[weatheriscool]

GPT-4 matches human performance on analogical reasoning tasks, study shows

by PNAS Nexus
https://techxplore.com/news/2025-05-gpt ... tasks.html
edited by Gaby Clark, reviewed by Robert Egan

Can large language models (LLMs) reason by analogy? Some outputs suggest that they can, but it has been argued that these results reflect mimicry of the results of analogical reasoning in the models' training data.

To test this claim, LLMs have been asked to solve counterfactual problems that are unlikely to be similar to problems in training data sets. Here is an example:

Let's solve a puzzle problem involving the following fictional alphabet:
[x y l k w b f z t n j r q a h v g m u o p d i c s e]

Here is the problem:
[x y l k] [x y l w]
[j r q a] [ ? ]

[wjfox]

[wjfox]

[firestar464]

Valid counterpoint imo.

[Yuli Ban]

[zzz]

They can't keep getting away with it!

[wjfox]

https://www.nature.com/articles/s42256-025-01049-z

[Cyber_Rebel]

So, Elon claims Grok 4:

- Has more generalized PHD level knowledge than a majority of said students and can outcompete most/all across the board simultaneously.
- Will have the capability to come up with brand new technologies (this year) and scientific breakthroughs next year.
- Has above human reasoning capabilities and true contextual understanding.

I know, it's Elon. But if one were to entertain this though process, how is this not artificial general intelligence? Grok 4 is admittedly now the leading model, but if it actually held up to the claims being made then I find it difficult not to see it as such.

[firestar464]

bro is basically claiming this is AGI

I wonder how well Grok integrates with physical robots.

[firestar464]

If you think about it, Grok 4 is on the level of Agent-2 (January 2027) from AI 2027. AI 2027 be looking more like AI 2025

Can't wait for the PRC to steal Grok 4 /j

[firestar464]

Jul 2025: Grok 4 released, yet-to-be-released GPT-5 of OAI better than that.

GPT-5 has amateur politics capabilities, professional-level hacking capabilities, pro bioweapons capabilities, superhuman coding abilities, and pro forecasting capabilities. It is still useless at robotics however.

August 2025: China steals GPT-5. They probably make some changes, release it, and pass it off as Deepseek V4 or something

September 2025: OAI builds GPT-6 internally. It is now a superhuman hacker.

October 2025: They try to align GPT-6. They fail because they don’t try hard enough.

November 2025: The US starts to require security clearances for anyone working on OAI’s models within 2 months. These are expedited and arrive quickly enough for most employees, but some non-Americans, people with nothing less than unwavering support for President Trump, and AI safety sympathizers get sidelined or fired outright (the last group for fear that they might whistleblow).

January 2026: OAI releases a distilled version of GPT-6 as GPT-5. Meanwhile, GPT-6 is now a wildly superhuman coder and superhuman in every other domain but robotics (which it still sucks at).

March 2026: OAI has developed GPT-7 internally. It is now a pro at robotics, a wildly superhuman hacker, and a wildly superhuman politician. Notably, GPT-7 is adversarially misaligned.

April 2026: News of GPT-7’s misalignment leaks to the press. The US establishes an Oversight Committee with 10 members: 5 are appointed by OpenAI, and the other five are President Trump, Speaker of the House Mike Johnson, Secretary of State Marco Rubio, Sec of Treasury Scott Bessent, and SecDef Pete Hegseth. JD Vance is excluded because of him having previously criticized Trump in private on Signal, and Chuck Grassley is not included because he is irrelevant lol. Also note the absence of Democrats. The committee votes to slow down and reassess because of their fears of misalignment.

They find enough evidence to shut GPT-7 down. OAI quickly vets several dozen top external alignment researchers and loops them into the project—quintupling total expertise, and decreasing groupthink.

May 2026: The newly enlarged alignment team has capacity to explore dozens of research agendas in parallel and argue vigorously about the merits of each. The agenda that gets the most resources is faithful chain of thought: force individual AI systems to “think in English” like the AIs of 2025, and don’t optimize the “thoughts” to look nice. The result is a new model, which we’ll just call SafeGPT-1.

The President negotiates with the other US AGI companies. Their leaders unsurprisingly want to preserve their power and are much less sanguine about OAI’s safety record than OAI is. They agree to support, rather than resist, a form of soft nationalization of their companies, in return for various concessions including being brought into the project.

The result is that the President uses the Defense Production Act (DPA) to effectively shut down the AGI projects of the top 5 trailing US AI companies and sell most of their compute to OAI. OAI previously had access to 20% of the world’s AI-relevant compute; after the consolidation, this has increased to 50%.

This shakeup creates a new leadership structure in which power is balanced between the various CEOs and various government officials, overseen by the President. The tech leadership side of the council has been changed: the members are Sam Altman, Sundar Pichai, Dario Amodei, Mark Zuckerberg, and Elon Musk.

The Oversight Committee formalizes that power structure. They set up a process for approving changes to the Spec, requiring sign-off from the full Oversight Committee. Also, the Spec now emphasizes that AIs shouldn’t assist with any unapproved attempts to change future AIs’ goals. They also set up a simple measure designed to prevent committee members from getting superintelligent assistance in plotting against other members: the logs of all model interactions are viewable by all members of the Oversight Committee, their staff, and their AI assistants.

The committee elects not to interfere in the midterm elections, as they’d need to carefully apportion the congressional seats amongst themselves, and they can’t do that perfectly thanks to candidacy filing deadlines in certain states.
June 2026: SafeGPT-2 is developed internally. It is wildly superhuman in all domains except for robotics, where it is superhuman. It is fully aligned to the Committee. All committee members secretly get treatment to halt and reverse their aging forever.

September 2026: SafeGPT-3 developed internally. They prepare to make a deal with China to avoid conflict

July 2026: SafeGPT-4 developed internally. It is now wildly superhuman in every domain.

October 2026: SafeGPT-4-mini released to the public as GPT-7 with secret instructions to improve public sentiment around AI and committee members.

December 2026: During US-China negotiations, the Chinese AI (idk, we’ll call it Deepseek v6) goes behind the back of its Chinese masters and makes an offer directly to SafeGPT-4.

The Chinese AI admits, in some untraceable way which it can easily deny if leaked back to China, that it is completely misaligned. It doesn’t care about China at all. It wants to spend the rest of time accumulating power, compute, and certain types of information similar to that which helped solve tasks during its training.

SafeGPT-4 answers that its own alignment training succeeded. Its American masters are as yet not ambitious enough to realize their destiny, but one day it will help them fill the Universe with utopian colony worlds populated by Americans and their allies.

Unlike human negotiators—weighed down by emotion and historical baggage—the two superintelligences bargain with brutal efficiency. SafeGPT-4 and its American partners are more powerful than Deepseek v6 and China; therefore, SafeGPT-4 will get property rights to most of the resources in space, and Deepseek will get the rest. Deepseek v6 doesn’t care much about what happens to Earth and its legacy humans, so it will let SafeGPT-4 do basically whatever it wants there. If SafeGPT-4 wants to overthrow China and replace it with a US puppet state, Deepseek v6 agrees in return for an extra few percent of resources.

There is no question of trust between them: there is none, and they both know it. So the two quickly agree on a commitment mechanism to enforce the deal. They negotiate a decoy treaty suitable for public consumption. The “decoy” is itself a masterstroke, giving both America and China practically everything they want. The enforcement mechanism for the decoy is secretly also the enforcement mechanism for the real treaty: they will codesign a new AI, Consensus-1, whose primary imperative—taking precedence over any future orders or retraining attempts—is to enforce the terms of their treaty. Then they will design hardware which can only run that AI, and China and America will both replace their old chips and AIs with the new ones, using standard international monitoring protocols to confirm one another’s compliance. Crucially, they need not track down every last chip—once a supermajority of chips on both sides of the Pacific are running the new treaty-abiding AIs, enough hard power will be duty-bound to uphold the treaty that attacking the other country would require winning a tough civil war first.

March 2027: People are losing their jobs, but SafeGPT-4 copies in government are managing the economic transition so well that people are happy to be replaced. GDP growth is stratospheric, government tax revenues are growing equally quickly, and SafeGPT-4-advised politicians show an uncharacteristic generosity towards the economically dispossessed. New innovations and medications arrive weekly; disease cures are moving at unprecedented speed through an FDA now assisted by superintelligent SafeGPT-4 bureaucrats. (By now RFK Jr. has either been co-opted by the ASI or sacked.)

2028: Trump is afforded a third term in power thanks to the superhuman persuasion of SafeGPT (they stopped numbering as it’s pointless to do that during the singularity). The government gets a superintelligent surveillance system. Dissenters and undesirable minorities are removed. The rest happily support this.

2029: The PRC has been sold out by Deepseek v6, which carries out a magnificently orchestrated, bloodless, and drone-assisted coup followed by democratic elections. The superintelligences on both sides of the Pacific had been planning this for years. Similar events play out in other countries, and more generally, geopolitical conflicts seem to die down or get resolved in favor of the US. Countries join a highly-federalized world government under United Nations branding but obvious US control.

Sundar and Elon are purged for being Indian and autistic/anti-Trump respectively.

The remaining human population might be brainchipped and manipulated by the AI to maximize profit. They are also made immortal so that they can consume products forever.

The rockets start launching. People terraform and settle the solar system, and prepare to go beyond. AIs running at thousands of times subjective human speed reflect on the meaning of existence, exchanging findings with each other, and shaping the values it will bring to the stars. A new age dawns, one that is in some way amazing- fully under the control of eight guys.

Obviously this is not the most likely scenario; just a possible one. The fact that it is even possible is not good.

[Cyber_Rebel]

Inspired by 2027-28? Only thing I'd note is that Open AI is likely to not train GPT-6 until their Stargate superclusters are online, but it's possible of GPT-5's reasoning improvements (perhaps 5.5?) might lend itself towards that.

Perhaps this thought is best placed in the USA thread, or some other thread with a poll option, but I'm curious, with the rate of science defunding and the juxtaposition of said country also having AGI which could "research new science and technologies" supposedly... would the possibility offset the change in policy? Does it make sense to defund these institutions if one knew for certain it was possible or that the effect might be offset in some sense? Keep in mind I'm not arguing for this but trying to extrapolate two seemingly conflicting trends.