Shawn

ML, Code, Networking. Hobbies: 20 years studying neuroscience, Guitar and tech.

[Shawn @Shawnryan96]

@shaunralston @_EdgeOfTheWeb @grok Have you tried advanced voice on grok? Is it good ?

[Shawn @Shawnryan96]

@nickhistgeek No we confabulate for many reasons from misfire to bad prediction. Not different root causes at all.

[Shawn @Shawnryan96]

Most of what we know is memorized and superficial, we are jagged in knowledge. We make many mistakes, we confabulate and hallucinate. I think all this will be persistent in AI, AGI and ASI even if we manage to mitigate a lot of it.

[Shawn @Shawnryan96]

@inductionheads @ylecun Anyone who thinks a model that learns patterns of patterns of patterns doesn't have some type of map of how it thinks things work Is in denial or has a poor understanding

[Shawn @Shawnryan96]

@MickWest There will always be a "roadblock."

[Shawn @Shawnryan96]

"They are taking our information and selling it back"
1. You put this data on the open web
2. It gets used for many things not just AI
3. It cost billions to train the models, which is why it cost $ to use
4. Unless you have tons of rich data your data is worth .0001 of cent

[Shawn @Shawnryan96]

@nickhistgeek I am not even sure domain knowledge is enough for physics. Most of us struggle beyond simple physics and we experience it. I think we are still a good ways off from solving some hard problems. (Not trying to sound like a downer)

[Shawn @Shawnryan96]

This is an experience issue. Imagine trying to create the look of a substance you have no experience with. The models just thinking text is gonna do it all is wrong. That part I agree with Yann about.

[Shawn @Shawnryan96]

@orionintx 90 percent of code looks just like LLMs. The average dev learns or copies from the internet. To get top end code it will need better data or continous learning and the ability to recall experience. It will be like this for all areas.

[Shawn @Shawnryan96]

@inductionheads I said long ago most of us will end up with basic models and they will lease out powerful model to those with certain credentials in bio, chem and cyber. They will use it as a high dollar subscription in the name of safety.

[Shawn @Shawnryan96]

@FutbolmeAI Code you would ship, however 90 percent of code isn't like that at all, though it should be. Like LLMs most people learn or copy from the internet and now we have a ton of crap code. LLMs need to be able to gain experience to get better

[Shawn @Shawnryan96]

@tunguz Comparing AI to crypto automatically marks you as an idiot.

[Shawn @Shawnryan96]

@shaunralston @AnthropicAI @Apple Can you imagine AI with Jobs still around? They may have not been the leader in AI but he would have pulled some magic from somewhere to make what they did have look like gold. I bet he is turning in his grave

[Shawn @Shawnryan96]

@debreuil I think we are limited by numbers. It took billions of us and thousands of years to get here. AI can work in the trillions at vast speeds. Ambition wont be its bottleneck but there will be one somewhere.

[Shawn @Shawnryan96]

@GregKamradt 10 yr old me would be freaking out that we actually have realistic conversations about this lol. Yes, they will need some type of a reward system.

[Shawn @Shawnryan96]

@GregKamradt The second would be ideal. Thats how humans work, but it has taken billions of us and thousands of years. In AI it could be instant but there are many security problems that would have to be solved.

[Shawn @Shawnryan96]

Even if you had advanced tech you would have to search billions of planets and still may not find anything like earth. Space is just too big

[Space Explained @SpacedExplained]

POV: Aliens looking for Earth in the universe

[Shawn @Shawnryan96]

@1337hero I am sure there is many ways to make money from AI that will seem obvious in hindsight. Building little apps isn't it though. That ship sailed many years ago.