Artificial General Intelligence (AGI) News and Discussions

[wjfox]

[spryfusion]

[spryfusion]

[wjfox]

[weatheriscool]

Weird how bs like religion is never questioned or pulsed but science and human advancement always is.

[spryfusion]

[firestar464]

1 year imo. FEEL THE AGI

[firestar464]

Weird how bs like religion is never questioned or pulsed but science and human advancement always is.

Well by then we'd have already achieved AGI for a few years, so at least that's some consolation.

[Yuli Ban]

Showerthought, and a summarization of David Shapiro's comment:
"AGI may already be here. We just haven't woken it up yet."

Based on Andrew Ng's post on agentic AI, and my own musings

GPT-3.5 (zero shot) was 48.1% correct. GPT-4 (zero shot) does better at 67.0%. However, the improvement from GPT-3.5 to GPT-4 is dwarfed by incorporating an iterative agent workflow. Indeed, wrapped in an agent loop, GPT-3.5 achieves up to 95.1%.

To me it seems plausible that a first-generation AGI ought to be a fusion of an LLM + agent swarm + concept search, and if that's the case, there's no reason why someone had not experimented to see if this is the case yet
In fact the way I think of it now is ironically like a corporation: foundational models are literally like the foundation, the building itself, which may have some functionality (e.g. utilities, internet connection, robotic security, etc.) But a corporation is not just a building (save for money laundering purposes). It's the humans inside that make it work, and they are a cascading hierarchy— agent swarms likely will operate similarly, with executive-level cognitive agents and adversarial agents that challenge decisions, and various smaller subagents with new ones "hired" by the larger ones as necessary.

[weatheriscool]

[weatheriscool]

[weatheriscool]

[Doozer]

In my opinion, AGI will be ideal in helping with creating and developing stories. I have made attempts with Chat GPT, but the results have often been frustrating and difficult given technical limitations. At its current level, the chatbot has very limited abilities regarding accuracy and consistency, hardly enough to satisfy my desires. I would like cite a few limitations AGI should overcome when implemented into the next generation of chatbots.

For example, when jumping from subject to subject, current modules can only memorize so much information before mixing things up and hallucinating. Within time, I am optimistic that AGI, should be able to at least reduce such a shortcoming, giving more accurate results. This would also allow discussions to flow more smoothly, without having to remind and correct the chatbot as frequently.

Another limitation currently facing Chat GPT is within its response mechanisms. All it can do now is spit what ideas you give it back at you while blindly praising them. What AGI needs is the ability to critique and expand on your ideas more easily, such as giving suggestions, providing more options or even pointing out flaws or inconsistencies in your ideas.

Lastly, AGI should be built with a better understanding of human emotions and desires. That way, it could know what readers and audiences want and provide suggestions to help make engaging stories and characters.

[wjfox]

Anthropic's Chief of Staff:

[weatheriscool]

AI Large Language Model Math Breakthroughs
June 16, 2024 by Brian Wang

AI large language models have been especially weak on math. There are now several papers from Google Deep Mind, Alibaba and other universities where AI large language models are at Math Olympiad levels and multiple step reasoning even with small models.

https://www.nextbigfuture.com/2024/06/a ... oughs.html

[spryfusion]

[spryfusion]

[weatheriscool]

OpenAI Internal Tracking to AGI Tiers
July 12, 2024 by Brian Wang

OpenAI reportedly internally introduced a new five-tier system to track its progress toward AGI. The classification system ranges from Level 1 (current conversational AI) to Level 5 (AI capable of running entire organizations)

https://www.nextbigfuture.com/2024/07/o ... tiers.html

[Cyber_Rebel]

Ah, so that's Open AI's version of the Google Deepmind's "Level's of AGI." It fits with what Sam Altman said regarding the measurement of AI capability rather than other metrics. Speaking of those capabilities, we're about to hit tier 2, 3, and possibly 4 seemingly quite soon:

Exclusive: OpenAI working on new reasoning technology under code name ‘Strawberry’

Reuters

July 12 - ChatGPT maker OpenAI is working on a novel approach to its artificial intelligence models in a project code-named “Strawberry,” according to a person familiar with the matter and internal documentation reviewed by Reuters.

The project, details of which have not been previously reported, comes as the Microsoft-backed startup races to show that the types of models it offers are capable of delivering advanced reasoning capabilities.

Teams inside OpenAI are working on Strawberry, according to a copy of a recent internal OpenAI document seen by Reuters in May. Reuters could not ascertain the precise date of the document, which details a plan for how OpenAI intends to use Strawberry to perform research. The source described the plan to Reuters as a work in progress. The news agency could not establish how close Strawberry is to being publicly available.

How Strawberry works is a tightly kept secret even within OpenAI, the person said.

Hopefully this could be the breakthrough needed to make it towards AGI.

[Yuli Ban]

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.