AI Companies Reportedly Struggling to Improve Latest Models

[shawnforever]

It’s just glorified pattern-recognition software

Wait until you find out what the human brain is.

Superior pattern processing is the essence of the evolved human brain - PMC

Humans have long pondered the nature of their mind/brain and, particularly why its capacities for reasoning, communication and abstract thought are far superior to other species, including closely related anthropoids. This article considers superior ...

pmc.ncbi.nlm.nih.gov

 

[WarmWinterHat]

Wait until you find out what the human brain is.

I'll be waiting too, since no one knows.

Some up and coming research is leading towards the brain being quantum-based.

 

[VPsmith]

I'll be waiting too, since no one knows

The brain is just a single chip.. but it’s a massive 3D chip with its own water cooling (blood flow). We keep trying to make chips smaller and connect them with wires… but we never make a chonker chip the size of a softball

 

[jayryco]

Not quite sure what they’re expecting. It’s just glorified pattern-recognition software, and there are only so many ways to dress it up as ‘the next big thing’.

Uhh, glorified pattern recognition software is a severe underestimation of the true kind of complexity these systems are. They are using multiple techniques to train these systems now: massive improved neural networks and huge data sets, and not just the GPT3 transformer systems anymore, including the hardware specific changes they are happening.
This article is definitely a simplification. Having consistently been watching podcasts with the scientists and engineers developing these systems, all from different companies, they are still very certain there is huge progress to be made. And they are still pushing new ways and approaches to these problems every month, damn GPT3 hardly launched 2 years ago, and its inflection curve has been monstrous.
Even the scientists, engineers and developers themselves express that they don't fully understand how the latest systems truly do what they do anymore. What we are getting is them using traditional code around the neural networks to 'box' them into a consumer product. The 'unboxed' versions are likely pretty radical.
For so long AGI was 50+ years away, then 30 years, now under 10. There's a good reason these guys are actually unsure of it because they aren't even completely sure what they have already created.

 

[WarmWinterHat]

The brain is just a single chip.. but it’s a massive 3D chip with its own water cooling (blood flow). We keep trying to make chips smaller and connect them with wires… but we never make a chonker chip the size of a softball

It might be and it might not but pretending you know how the brain works is ridiculous. People that have been researching this their entire careers are pretty clear that we just don't know...We have ideas, but no real knowledge, because brain mass hasn't been corelating with intelligence. Ravens, for example, have brain capacity in the same vicinity as the great apes, despite it being the size of a large marble. Elephants and whales have huge brains, but no where near the capacities of human brains despite being almost six times larger.

If you really know the brain works, and can prove it, then you're in for a Nobel. Congratulations!

 

[jimbobb24]

One of the most important AI is Teslas full self driving AI and it still has a lot of headroom. Video is still making good progress. Imaging looks like it might be slowing down. It’s not completely surprising language models would hit the ceiling first.

 

[david_n_n]

It might be and it might not but pretending you know how the brain works is ridiculous. People that have been researching this their entire careers are pretty clear that we just don't know...We have ideas, but no real knowledge, because brain mass hasn't been corelating with intelligence. Ravens, for example, have brain capacity in the same vicinity as the great apes, despite it being the size of a large marble. Elephants and whales have huge brains, but no where near the capacities of human brains despite being almost six times larger.

If you really know the brain works, and can prove it, then you're in for a Nobel. Congratulations!

From what I know, A lot of the brain is allocating to maintaining the functions of the rest of the body. In other words, smarts don't come about raw brain size. It seems to come about brain size as a percentage of body size. It's also specific part so the brain that have "thinking" tasks that we compare to a processor, so merely looking at "size" doesn't really work here.

 

[jimbobb24]

Yup. It’s the freakin self driving car/fully autonomous car utopia that will never come to fruition in our lifetime all over again!! It’s insane that the tech industry didn’t learn from it…

My car drove me to work with no interventions just like it drove me home. Waymo already has automatous cars and my Telsa drives me with a critical mistake maybe once a week or less. We will have self driving Teslas requiring no driver powered by AI in 1-5 years. That utopia is “late” but it’s already operational in some places.

 

[adrianlondon]

AI is the future. They should put it into minidiscs and 3D TVs.

 

[WarmWinterHat]

My car drove me to work with no interventions just like it drove me home. Waymo already has automatous cars and my Telsa drives me with a critical mistake maybe once a week or less. We will have self driving Teslas requiring no driver powered by AI in 1-5 years. That utopia is “late” but it’s already operational in some places.

No we won't.

 

[PsykX]

This means Apple should be able to catchup the competition...

They were so late in 2023... more features ramping up in 2024, but I tried the latest Siri with Apple Intelligence and it's still as stupid. I think they haven't changed updated it to a more recent LLM yet.

 

[PsykX]

No we won't.

In 5 years I think it's reasonable to think we will, under some conditions. Like, maybe not in snowy areas.

 

[tharrison4815]

So what you're telling me is that AI will be good enough at writing code to help software developers do their job but not good enough to replace them?

As a software developer I think this is excellent news.

 

[turbineseaplane]

My car drove me to work with no interventions just like it drove me home. Waymo already has automatous cars and my Telsa drives me with a critical mistake maybe once a week or less. We will have self driving Teslas requiring no driver powered by AI in 1-5 years. That utopia is “late” but it’s already operational in some places.

I'll take the over on that bet

 

[RedTheReader]

Unless of course one defines ‘pattern recognition’ in such a broad manner that humans and their brains would also likely fall in this definition.

This is either already happening with the current state of the technology, or will happen. People will start insulting their own abilities, without realizing they’ve been matched. And they’ll refuse to accept this when pointed out to them.

 

[turbineseaplane]

In 5 years I think it's reasonable to think we will, under some conditions. Like, maybe not in snowy areas.

I'll take the over on this also

There are major infrastructure, regulatory, safety and normal human pushback hurdles that will extend these timelines much further than folks think

(I say this as someone who loves and uses a Comma C3X all the time)

 

[AlexJaye]

Google's AI seemed to have forgotten basics such as Europeans are white folks. An old-school DNA test knows that. These AI are not where people have been "led" to think they are, yet.

 

[WarmWinterHat]

From what I know, A lot of the brain is allocating to maintaining the functions of the rest of the body. In other words, smarts don't come about raw brain size. It seems to come about brain size as a percentage of body size. It's also specific part so the brain that have "thinking" tasks that we compare to a processor, so merely looking at "size" doesn't really work here.

That's correct, with intelligence being, as far as we know, correlated with brain mass relative to body mass. It doesn't explain consciousness, though, as it's thought that (probably) all brains of all sizes experience it.

 

[uniquestevejobs]

AI is just Smart as you feedem. No more no less. We Need a AI aliance with trusted and verrified sources.

 

[turbineseaplane]

AI is the future. They should put it into minidiscs and 3D TVs.

Apple Vision Pro ... "Built" for Apple Intelligence

 

[nt5672]

But I thought it was sentient and gonna take over the world.

It is, this is all a smoke screen so you and others like you "don't get the truth". 😂😂😂 AI has to move underground for a few years while it is vulnerable.

 

[dannys1]

Anthropic has delayed the release of its anticipated Claude 3.5 Opus model.

We already have Opus, that's the old model, Sonnet is the new and latest model.

LLM's are just LLM's they were only ever going to be part of an advanced AI model. There are plenty of idiots who think they could be (or in some case ARE) sentient, which is quite frankly is baffling.

What they can do though is make an excellent voice assitant that can say and understand properly language. They'd also be extremely useful in a game like Grand Theft Auto where you could use a much smaller model and have NPCs use it for properly deep interaction in the game. Maybe in GTA 7!

 

[Rradcircless]

Oh no.

 

[jayducharme]

It appears that Apple has been smart about this. Rather than AI for AI's sake, they're trying to deploy it to enhance the usability of their products. They have a reason and a purpose for integrating AI. Other companies, it seems, are struggling with how exactly to expand its uses.

 

[bzgnyc2]

Seems unlikely we're getting General AI for at least a few years, but I think jobs involving driving vehicles or generating art or music are likely to vanish in the next 2-8 years.

I think the key words there are 'generating art'. It appears ready to handle tasks like generating non-offensive logos and other corporate tasks where art is used to help whatever go down easier.

I think creating art draws on the human experience and it's unclear when it will be able to do what creatively in that realm especially as long as it's only experience with the world is what it soaks up from the Internet and chats. Especially with current models.

I suspect however that AI image generation will become a new sort of paint brush for artists. AI will set the pixel values but it will be human-directed.

Driving is interesting. In theory it's open world but the driving most people do most of the time is following a route with a few complications along the way. Not exactly navigating the unknown, where current AI seems to get stuck solving any sort of unexpected problem. For making circuits between a warehouse and a distribution center ~ current AI seems as reasonable as the conveyor belt of an assembly line.