Not liking Ai

[Gants-P]

No one said Sam Altman isn't selling out ;-)

Ilya is definitely not Sam.

Sam is not a scientists AT ALL.

He's a business nerd/tech guy who can code. He isn't motivated by science. Ilya is motivated by science.

Sam is a MASSIVE sell-out. Not the guy to be leading up AI, but he's useful to those in power because of the way he is.

To be more precise though, Sam is motivated by power. He's a bit of a dark character in my opinion.

At that level no one is motivated by money.

People are either motivated by the power money gives you, or they're motivated by what they can create with money. People are always one of the 2.

Elon is a creation guy.

Sam is a power guy.

Ilya is a creation guy, but he's a pure scientist, so he doesn't directly involve himself in money.

Elon is a hybrid scientist/business person. About 30% science, 70% business.

In 2024 and beyond the hybrids will make the most money. In the 50s to 80s it was the pure business/money/negotiation guys like Trump.

Sam is a hybrid too actually, but one motivated by power, not creation. He definitely understands the tech. That's what makes Sam more dangerous. He looks fluffy on the outside, but on the inside he craves power *and* fame.

You do realize there are other people besides illya sutskever, just because he's trending doesn't mean you need to mention only he's name.

But you are astute on the power dynamics I'll give you credit there.

 

[splishsplash]

You do realize there are other people besides illya sutskever, just because he's trending doesn't mean you need to mention only he's name.

But you are astute on the power dynamics I'll give you credit there.

What would you like me to do..

Deep research into every AI engineer on the planet along with a full analysis of their career and motives?

Ilya and Sam were mentioned which is why I was talking about him. I brought up Elon because he's an example of a hybrid that's motivated by creation vs Sam the hybrid who is motivated by power.

Specifically one of the previous posters said Ilya was a sellout, so that's why I was analysing him, as I really don't think he's a sellout at all. He's just going where he's going to have the most fun building cool shit.

 

[Skinwalter]

they are just flooding the internet with automated crap

 

[noellarkin]

There's definitely a hype cycle in AI - - this is primarily a byproduct of marketing that coaxes investors to pay more attention to AI startups. This doesn't mean that AI is useless or "unlikeable". I'll assume that by AI you're talking about LLMs.

The best use case I've found for GPT models is RAG, which is LLM + querying a database for "grounding" the model in facts or internal company documentation. This isn't easy to do, takes a fair amount of work to prevent semantic drift and determine how much grounding information is appropriate...but good RAG pipelines + LLMs are something you'll want to check out, if the limitations of massive GPT models annoys you.

The other significant development has been tiny but highly performant models - - I love how good Phi-3 is, for example, especially in smaller contexts.

If you're upset that an AI tool can't make Apple or Android apps, at one level it's understandable, because many of these companies have been over-selling the abilities of their products, at another level it also means that you may need to look into how these models work, and although they can't do the work for you, you can probably find ways to use them to make you more efficient when you're coding your app.

Regarding the deterioration of the SERPs because of ChatGPT content, that does annoy me, but then I remind myself that before ChatGPT, we had deterioration of the SERPs due to spintax, which was pretty bad too. The writing style and tone that ChatGPT has been trained to use does grate on the nerves a little (especially the refusals) but IMO that itself is the result of anthropomizing it. If you treat it like a tool instead of a person, it's a lot easier to deal with.

SORA and other similar tools (Runway etc) will be useful for creating B-reels and stock footage. Don't expect it to create a movie from scratch, but it will be very useful for Youtube creators.

At a strategic level, LLMs are an amazing tool for generalists - - an LLM won't make you an expert in anything, but it cuts your learning curve by reducing the amount of research you have to do. Example: get 20 ebooks on your subject matter, throw them into ChromaDB or a vector store of your choice, slap an LLM in front of it and now you can "chat" with your books. So many problems can be reduced to search and sort problems, and I think of vector search as a great way to augment old-school search (TF-IDF and all that).

Just manage your expectations, and you'll find good use cases for LLMs. Don't expect them to "reason" they're not really built for that (I've kept an open mind about this, but after a year and a half of working with these things, IMO they simulate reasoning with token prediction but don't actually reason - - system 1 that imitates system 2) - - they're "just" prediction systems, but prediction systems can be extremely useful.

 

[rusomike90]

There's definitely a hype cycle in AI - - this is primarily a byproduct of marketing that coaxes investors to pay more attention to AI startups. This doesn't mean that AI is useless or "unlikeable". I'll assume that by AI you're talking about LLMs.

The best use case I've found for GPT models is RAG, which is LLM + querying a database for "grounding" the model in facts or internal company documentation. This isn't easy to do, takes a fair amount of work to prevent semantic drift and determine how much grounding information is appropriate...but good RAG pipelines + LLMs are something you'll want to check out, if the limitations of massive GPT models annoys you.

The other significant development has been tiny but highly performant models - - I love how good Phi-3 is, for example, especially in smaller contexts.

If you're upset that an AI tool can't make Apple or Android apps, at one level it's understandable, because many of these companies have been over-selling the abilities of their products, at another level it also means that you may need to look into how these models work, and although they can't do the work for you, you can probably find ways to use them to make you more efficient when you're coding your app.

Regarding the deterioration of the SERPs because of ChatGPT content, that does annoy me, but then I remind myself that before ChatGPT, we had deterioration of the SERPs due to spintax, which was pretty bad too. The writing style and tone that ChatGPT has been trained to use does grate on the nerves a little (especially the refusals) but IMO that itself is the result of anthropomizing it. If you treat it like a tool instead of a person, it's a lot easier to deal with.

SORA and other similar tools (Runway etc) will be useful for creating B-reels and stock footage. Don't expect it to create a movie from scratch, but it will be very useful for Youtube creators.

At a strategic level, LLMs are an amazing tool for generalists - - an LLM won't make you an expert in anything, but it cuts your learning curve by reducing the amount of research you have to do. Example: get 20 ebooks on your subject matter, throw them into ChromaDB or a vector store of your choice, slap an LLM in front of it and now you can "chat" with your books. So many problems can be reduced to search and sort problems, and I think of vector search as a great way to augment old-school search (TF-IDF and all that).

Just manage your expectations, and you'll find good use cases for LLMs. Don't expect them to "reason" they're not really built for that (I've kept an open mind about this, but after a year and a half of working with these things, IMO they simulate reasoning with token prediction but don't actually reason - - system 1 that imitates system 2) - - they're "just" prediction systems, but prediction systems can be extremely useful.

What a answer, well said dude. Everything agreed

 

[Nikolaos]

They got heavy backlash from the NSA director hiring and lost one of their top engineers Ilya Sutskever who went to start his own company. Curious to see what he does differently. We are using overhyped products. Much more powerful marketing tools could be released but still discussing usage costs & limits. I would like for a gpt to be released by Openai that could make apple apps and android apps without a issue not the knock off made versions, I mean one that would go mainstream how good it is. It exists, but limited use.

In order to make apps it has to be trained very well! I don't think we have reached that point with A.I. yet...

 

[abbesnetwork]

AI is good but there are precautions in using AI. You need also to include your ideas and rephrase each sentences to have authenticity on your work.