Researchers slam OpenAI’s GPT-4 paper as ‘masquerading as research’
There is a lot of criticism right now about the newly released GPT-4 research paper. What are the biggest issues?
William Falcon: I think what’s bothering everyone is that OpenAI made a whole paper that’s like 90-something pages long. That makes it feel like it’s open-source and academic, but it’s not. They describe literally nothing in there. When an academic paper says benchmarks, it says ‘Hey, we did better than this and here’s a way for you to validate that.’ There’s no way to validate that here.
Is this different than it was when ChatGPT came out? Or DALL-E? Were those masquerading as research in the same way?
Falcon: No, they weren’t. Remember, GPT-4 is based on Transformer architecture that was open-sourced for many years by Google. So we all know that that’s exactly what they’re using. They usually had code to verify. It wasn’t fully replicable, but you could make it happen if you knew what you’re doing. With GPT-4, you can’t do it.
So would you say that this step goes farther than anything from Google, or Microsoft, or Meta?
Falcon: I think [this is] the first time I’m seeing this is not possible [to replicate], based on this paper.
There is a lot of criticism right now about the newly released GPT-4 research paper. What are the biggest issues?
William Falcon: I think what’s bothering everyone is that OpenAI made a whole paper that’s like 90-something pages long. That makes it feel like it’s open-source and academic, but it’s not. They describe literally nothing in there. When an academic paper says benchmarks, it says ‘Hey, we did better than this and here’s a way for you to validate that.’ There’s no way to validate that here.
Is this different than it was when ChatGPT came out? Or DALL-E? Were those masquerading as research in the same way?
Falcon: No, they weren’t. Remember, GPT-4 is based on Transformer architecture that was open-sourced for many years by Google. So we all know that that’s exactly what they’re using. They usually had code to verify. It wasn’t fully replicable, but you could make it happen if you knew what you’re doing. With GPT-4, you can’t do it.
So would you say that this step goes farther than anything from Google, or Microsoft, or Meta?
Falcon: I think [this is] the first time I’m seeing this is not possible [to replicate], based on this paper.
👍11❤2
Bing with ChatGPT: Using chat to do work using web page content
User: hey can you calculate the total interest payed?
Bing: Sure! According to the information provided on the web page context, the total interest paid over 12 months is $1,172.22. This is calculated by adding up all the interest paid for each payment: $172.72 + $159.28 + $145.66 + $131.88 + $117.93 + $103.81 + $89.52 + $75.05 + $60.40 + $45.574 + $30.574 + $15.38 = $1,172.22.
User: hey can you calculate the total interest payed?
Bing: Sure! According to the information provided on the web page context, the total interest paid over 12 months is $1,172.22. This is calculated by adding up all the interest paid for each payment: $172.72 + $159.28 + $145.66 + $131.88 + $117.93 + $103.81 + $89.52 + $75.05 + $60.40 + $45.574 + $30.574 + $15.38 = $1,172.22.
👍14❤1
GPT-4 can create a "rough approximation of a mango" using FreeCAD macro scripting in Python
“I asked GPT-4 to create a rough approximation of a mango using FreeCAD.
It generated a 2-dimensional oval connected to another oval, with some sort of torus shape. I asked it to add 100 layers of ovals to approximate the shape. It did the same exact torus shape, but more smoothly.
I told it there's a torus in the middle and not an obtrusion as a mango might have. It altered the torus shape to produce this glorious specimen.”
“I asked GPT-4 to create a rough approximation of a mango using FreeCAD.
It generated a 2-dimensional oval connected to another oval, with some sort of torus shape. I asked it to add 100 layers of ovals to approximate the shape. It did the same exact torus shape, but more smoothly.
I told it there's a torus in the middle and not an obtrusion as a mango might have. It altered the torus shape to produce this glorious specimen.”
❤1
This media is not supported in your browser
VIEW IN TELEGRAM
ChatGPT GPT4 plays chess against Stockfish 15.1 (Stockfish is White)
🎄23🔥16❤4👍3
GPT-4 Does the DEI Die List
Thrilled for day when GPT’s wokeness abilities cross so far into super-human level, that both sides end up doing all their wokeness generation and evaluation using GPTs, so that it ends up just each side’s GPTs talking to each other.
Thrilled for day when GPT’s wokeness abilities cross so far into super-human level, that both sides end up doing all their wokeness generation and evaluation using GPTs, so that it ends up just each side’s GPTs talking to each other.
😱13❤2🤡2
AI Alignment Obstacle: 61% identifying as “right-wing” in a group say they agree with “fictitious capital” belief
I.e. the cornerstone of belief of Marx’s 1894 book, foundational belief of numerous communist uprisings through out history, and fundamentally inconsistent with the core right-wing beliefs, on the primary dimension separating right from left beliefs.
Similar inconsistencies seen among those identifying as left.
So, how then are we supposed to align AI’s with our beliefs — when we can’t even consistently align our beliefs with themselves?
AI alignment problem.
I.e. the cornerstone of belief of Marx’s 1894 book, foundational belief of numerous communist uprisings through out history, and fundamentally inconsistent with the core right-wing beliefs, on the primary dimension separating right from left beliefs.
Similar inconsistencies seen among those identifying as left.
So, how then are we supposed to align AI’s with our beliefs — when we can’t even consistently align our beliefs with themselves?
AI alignment problem.
🤡10👍7❤3🤨3👏2