The Singularity - Official Thread

Talk about scientific and technological developments in the future
User avatar
wjfox
Site Admin
Posts: 8732
Joined: Sat May 15, 2021 6:09 pm
Location: London, UK
Contact:

Re: The Singularity - Official Thread

Post by wjfox »

Wow, we just exceeded the 100 billion mark for transistor counts...

https://en.wikipedia.org/wiki/Transistor_count

https://www.apple.com/newsroom/2022/03/ ... -computer/

Moore's Law is still on track. 8-)
User avatar
wjfox
Site Admin
Posts: 8732
Joined: Sat May 15, 2021 6:09 pm
Location: London, UK
Contact:

Re: The Singularity - Official Thread

Post by wjfox »

Transistor gate is just 0.3 nm long

20th March 2022

Using graphene and molybdenum disulphide, scientists in China have made a transistor gate with a length of only 0.3 nanometres, equivalent to just one carbon atom, by exploiting the vertical aspect of the device.

[...]

"We have realised the world's smallest gate-length transistor," said Tian-Ling Ren, senior study author and an electrical engineer at Tsinghua University in Beijing. "The next goal is making a 1-bit CPU."

Read more: https://www.futuretimeline.net/blog/202 ... -china.htm


Image
User avatar
Ozzie guy
Posts: 486
Joined: Sun May 16, 2021 4:40 pm

Re: The Singularity - Official Thread

Post by Ozzie guy »

Of late I have been thinking "Human level AI" is a very vague concept, there are thousands of possible combinations of an AI being better than humans at a lot of things whist worse than humans at a lot of things.

Proto AGI is actually a better definitive line in the sand, you ether have very retarded AGI or you don't.

What is nice is that proto AGI is believed to be at the latest only 2 years and 9 months away and will mark the start of AGIs getting better at an exponential rate.

I guess this is only relevant psychologically as a change in morale but I will drop waiting for the vague human level AI and wait for proto AGI like I did in the past which is much closer.
TrueAnimationFan
Posts: 119
Joined: Wed May 19, 2021 8:00 pm

Re: The Singularity - Official Thread

Post by TrueAnimationFan »

Set and Meet Goals wrote: Sun Mar 27, 2022 1:47 am Of late I have been thinking "Human level AI" is a very vague concept, there are thousands of possible combinations of an AI being better than humans at a lot of things whist worse than humans at a lot of things.

Proto AGI is actually a better definitive line in the sand, you ether have very retarded AGI or you don't.

What is nice is that proto AGI is believed to be at the latest only 2 years and 9 months away and will mark the start of AGIs getting better at an exponential rate.

I guess this is only relevant psychologically as a change in morale but I will drop waiting for the vague human level AI and wait for proto AGI like I did in the past which is much closer.
I always assumed when people said "human level AI" they meant something that you'd expect more like 20 years from now, where there may be a (brief?) stretch of time when man and machine are basically at the same level of skill, thought speed and communication ability.
Last edited by TrueAnimationFan on Mon Mar 28, 2022 5:01 pm, edited 1 time in total.
User avatar
Yuli Ban
Posts: 4631
Joined: Sun May 16, 2021 4:44 pm

Re: The Singularity - Official Thread

Post by Yuli Ban »

The concept of "proto-AGI" stems directly from my early musings about a "weak AGI"

In other words, around 2015 or so, I noticed that our delineation of AI abilities was much too muddled. We had multiple terms to describe the same thing, so I decided to reuse "weak" and "narrow"/"strong" and "general" to have different meanings.
Weak AI doesn't necessarily mean narrow/single-purpose/brittle AI. It just means any AI that's not par-human in ability. So it's possible to have a "strong narrow AI."
Strong AI doesn't necessarily mean general/multi-purpose/complete AI. It just means any AI that's above human in ability. So it's possible to have a "weak general AI."

I also realized that there had to be an intermediate stage of AI, something in between narrow and general AI.

Of course I have zero sway in the field of AI for obvious reasons, so it makes sense this never took off. But I'm still fascinated by that twilight range, and at some point I figured that it'd simply be easier to fall back on "proto-AGI" to describe those edge-cases. Surely there has to be a point where we develop multipurpose AI, and it's not going to be so strict as "today, we only have narrow AI, tomorrow we have human-level general AI." We already see this fuzziness with general-learning algorithms and transformers. Hence my reasoning that proto-AGI must be possible, that we will develop a multipurpose tool that won't be sentient or human-level, but will allow us to generalize across all tasks.
And remember my friend, future events such as these will affect you in the future
User avatar
Yuli Ban
Posts: 4631
Joined: Sun May 16, 2021 4:44 pm

Re: The Singularity - Official Thread

Post by Yuli Ban »

And remember my friend, future events such as these will affect you in the future
Tadasuke
Posts: 510
Joined: Tue Aug 17, 2021 3:15 pm
Location: Europe

Re: The Singularity - Official Thread

Post by Tadasuke »

As time goes on, I believe in the Singularity less and less. But not because I think the future is worse than the present, or that the future won't bring exponential progress, but because I'm not sure if exponential progress must mean that the Singularity happening. I do think that there will come a time when some people do say that the Singularity is happening now, but not everyone will agree. If the Singularity will happen, I predict it in 2114. This is based upon tech acceleration. The Singularity is when a single augmented human being or a single AI becomes more intelligent and more capable (in everything) than the entire human race combined. I don't think that's close. Doesn't look to be close. Unless you call 2114 as close. But yes, progress is definitely exponential.
Global economy doubles in product every 15-20 years. Computer performance at a constant price doubles nowadays every 4 years on average. Livestock-as-food will globally stop being a thing by ~2050 (precision fermentation and more). Human stupidity, pride and depravity are the biggest problems of our world.
User avatar
Yuli Ban
Posts: 4631
Joined: Sun May 16, 2021 4:44 pm

Re: The Singularity - Official Thread

Post by Yuli Ban »

And remember my friend, future events such as these will affect you in the future
User avatar
Yuli Ban
Posts: 4631
Joined: Sun May 16, 2021 4:44 pm

Re: The Singularity - Official Thread

Post by Yuli Ban »

And remember my friend, future events such as these will affect you in the future
User avatar
Yuli Ban
Posts: 4631
Joined: Sun May 16, 2021 4:44 pm

Re: The Singularity - Official Thread

Post by Yuli Ban »

starspawn0
It surpasses Chinchilla; and, they applied chain-of-thought reasoning to make progress on math reasoning. Still not off-the-charts strong on math -- but definitely improving.

Addendum: The improvement on GSM8K is a lot bigger than mentioned in the "self-consistency" paper:

https://arxiv.org/abs/2203.11171

In that paper, the top score mentioned was 35% accurate using fine-tuning on 7,500 examples + access to calculator. But then there was a score of 55% achieved using a fine-tuned GPT-3; and, now, Pathways brings that up to 58%.

The Pathways blog post makes no mention of "self-consistency", though maybe that's what they mean by "8-shot prompting" (I don't think so -- I think that's just the 8 prompts they used in all their previous papers to trigger models to answer math questions accurately). Did they make use of it? If not, it should lead to further gains if used -- maybe pushing performance well past the 60% threshold they mention.
And remember my friend, future events such as these will affect you in the future
Post Reply