Over for Google and OpenAI

[Deleted member 131140]

‘A complete accuracy collapse’: Apple throws cold water on the potential of AI reasoning – and it's a huge blow for the likes of OpenAI, Google, and Anthropic

Presented with complex logic puzzles, AI reasoning models simply gave up

www.itpro.com

Apparently, the researchers say that the reasoning models have 0 accuracy as the logical reasoning tests go up in complexity

The ques for the established benchmarks might already have answers baked into the training set of the models so they were inaccurate when assessing a model's accuracy

@Jason Voorhees career extended by 20 years

Link to paper

https://ml-site.cdn-apple.com/papers/the-illusion-of-thinking.pdf

 

[Deleted member 131140]

@Snicket

 

[Deleted member 131140]

@Chimera

 

[DirtyBlonde]

chatgpt couldn't understand my 1 sentence prompt today, can confirm

 

[Deleted member 131140]

chatgpt couldn't understand my 1 sentence prompt today, can confirm

Its for purely logical reasoning tasks though.

Specifically highly complex tasks

 

[Deleted member 117288]

exactly, AI is extremely gay and over-hyped to the point where it cringes you

 

[Deleted member 131140]

exactly, AI is extremely gay and over-hyped to the point where it cringes you

Glad you fixed your retarded cellphone

 

[Deleted member 131140]

Bump

 

[Deleted member 117288]

Glad you fixed your retarded cellphone

Nah I didn't bother with fixing that shit I just threw it away.

pcs are much more convenient, metaphysically

 

[Deleted member 158882]

Can’t be bothered with this tech stuff.

Just going outside and doing anything mogs.

 

[Gonthar]

Lol, Apple has been left behind in the AI department, that's why they desperately try to downplay the importance of AI...

 

[Deleted member 131140]

Lol, Apple has been left behind in the AI department, that's why they desperately try to downplay the importance of AI...

The problems in the paper were not very tough though. The river problem was simple enough for humans to do it

 

[Deleted member 91663]

@Snicket

I haven't been following the story very closely.
What's the bottleneck of AI in personal phone usage?
Surely some kind of GPT based model could do an adequate job? But obviously it's more complex than that.

 

[piec]

this is why grok mogs

 

[Lord Shadow]

obviously lol for anyone thinking AI wasnt going to be any different from just a smarter google search than its over @lifeless

 

[Deleted member 131140]

I haven't been following the story very closely.
What's the bottleneck of AI in personal phone usage?
Surely some kind of GPT based model could do an adequate job? But obviously it's more complex than that.

Its not personal usage to be specific.

The reasoning models just fail when given deterministic algorithmic problems.

It should be doable for them since its within the context window but they just "give up" before attempting as more variables are stacked on

 

[Basedman420]

because Ai is stupid they cant think

 

[Basedman420]

because Ai is stupid they cant think

i just realised i typed "they'

 

[Bitchwhipper2]

Chatgpt is a genuine retard tbh

 

[Deleted member 131140]

Chatgpt is a genuine retard tbh

Still has a good knowledge base though. 2030 will be either complete bust and start an AI winter or a complete boom

 

[gpsl]

Tell that to the soldiers who get their faces blasted off by machine learning drones

 

[Deleted member 131140]

Tell that to the soldiers who get their faces blasted off by machine learning drones

You dont need high level reasoning for that.

A bqsic Coordinate system and tracer technology can do that

 

[Bitchwhipper2]

Still has a good knowledge base though. 2030 will be either complete bust and start an AI winter or a complete boom

Yea, it has a good bit of knowledge to draw upon.

Good data analytics. But asking for its own take on philosophical ponderings is just shooting yourself in the foot

 

[gpsl]

You dont need high level reasoning for that.

A bqsic Coordinate system and tracer technology can do that

Openai should hire the basement chinese to fix their retarded models

 

[Gonthar]

I haven't been following the story very closely.
What's the bottleneck of AI in personal phone usage?
Surely some kind of GPT based model could do an adequate job? But obviously it's more complex than that.

They can't run the top LLM models locally, you would need a very powerful computer and lots of RAM for that, mostly they are run in the cloud on special servers.

 

[Deleted member 91663]

They can't run the top LLM models locally, you would need a very powerful computer and lots of RAM for that, mostly they are run in the cloud on special servers.

Good point. Hadn't considered this.
Why can’t phones have cloud-based AI instead of on-device?

 

[Deleted member 91663]

Its not personal usage to be specific.

The reasoning models just fail when given deterministic algorithmic problems.

It should be doable for them since its within the context window but they just "give up" before attempting as more variables are stacked on

Interesting.

Is Apple limited by trying to develop on-device AI versus using something server-based like Chat GPT or Gemini instead?

 

[Gonthar]

Good point. Hadn't considered this.
Why can’t phones have cloud-based AI instead of on-device?

It still takes a few seconds until you get a response from ChatGPT, cloud performance can fluctuate a lot depending on how many users are online and using that service, or the Internet speed, etc., a phone would simply lag too much if you would have to wait for seconds for a response to your various requests.

 

[Deleted member 91663]

It still takes a few seconds until you get a response from ChatGPT, cloud performance can fluctuate a lot depending on how many users are online and using that service, or the Internet speed, etc., a phone would simply lag too much if you would have to wait for seconds for a response to your various requests.

Yeah, makes sense. And with over 1 billion iPhone users worldwide, the cloud infrastructure costs would be astronomical.