The Surprising Outcome of This Challenge

[@DataRae-AIEngineer]

Become an expert on trending AI topics with my monthly newsletter: datarae.ck.page/0dac61784d

[@Gilb1037]

id recommend reducing music volume in some areas, we want to hear you more than the music

[@RickeyBowers]

FYI: It didn't go crazy - you need to start a new session, or it will assume the conversation is continuing. I think the models are useful. Thanks for sharing your discoveries.

[@cutemartinj]

Love your energy!

[@minneso5424]

Thanks much! Rae

[@mradford10]

It slows down because you keep each new question in the same session. Create a new chat for each new question and watch it scream. It’s processing the ENTIRE history of your chat every time you add another question… so that in itself is compelling and not that slow at all.

[@milque1854]

flash attn cntx shifting and RAG all exist and are definitely being used in gpt4

[@hebozhe]

I know for a fact it can't do college-level logic -- syntactic derivations or model theory. I suspect it's something about the novelty and length of our proofs (i.e., not enough examples of proofs, semantic tableau, etc.). Like, it doesn't really get inference rules, just kind of guesses at them. Also, it has no clue how to properly differentiate classical proofs from intuitionistic ones, minimal ones, ones from particular modal systems, etc.

[@DataRae-AIEngineer]

Thanks for commenting. Yea I want to play with its math skills a little more. TBH I am an applied math person, not a logician. It would be fun to see a math channel do that though. You should post it :)

[@NorbertKasko]

6:35 I was entirely sure it's gonna say: Sure! It never says "of course" for instance.

[@DataRae-AIEngineer]

yea I wasn't sure what was going on there. I wish we could get it to be sarcastic lol.

[@NorbertKasko]

@@DataRae-AIEngineer It also never says! "Yes, I can do it!" or something like that. It always just "sure" it can do things. It's a bit boring.

[@KonChEkey]

The scariest thing about agi is it might already exist, just pretending to be unthinking & wait till humans are totally dependent on it

[@DataRae-AIEngineer]

Interesting thought.

[@vincehomoki1612]

This is the first time I've seen your channel, what is qstar?

[@DataRae-AIEngineer]

Hi. Thank you so much for visiting the channel! Q-Star is a controversial potentially hypothetical / probably real algorithm alleged to be created by OpenAI. In November, their CEO was temporarily fired by the board because they said this Q-Star thing was unsafe and a threat to humanity because of how innovative and similar to human-thinking it was. Microsoft ended up flexing their muscles becuase OpenAI uses Microsoft Azure, and the CEO of OpenAI ended up returning and the board members who voted him out were fired. It's a big drama in the computer science space, ongoing even now.

[@whemmakatatt5311]

Loving the content, keep em coming :D

[@RomboDawg]

that fucking intro went hard

[@DataRae-AIEngineer]

Awe thanks I worked really hard on it

[@jmirodg7094]

This is great, but look at open sources capabilities, with the various agent frameworks you can do amazing things and will not be snooped and handcuffed by (not so) openAI

[@DataRae-AIEngineer]

Thank you. I'm not wed to any specific AI framework. Which ones do you like? Maybe I'll do a video while I play with them.

[@jmirodg7094]

@@DataRae-AIEngineer I tried Crew AI, easy to use with different llm inluding Groq API and local Llama3, it is working amazing on some stufs but it has it's quirks. Starting to try Autogen 2. Some time ago I started to develop my own kolergy/agent_bean on github but paused for now to see if Crew or autogen can do what i want

[@jmirodg7094]

​@@DataRae-AIEngineer I tried Crew AI which works but have it's quirks, i need to try Autogen i had been developing my own one (kolergy agent_bean) but I paused it to see if those could do the job

[@jmirodg7094]

i tried crewAI with some success and i need to try autogen

[@jmirodg7094]

@@DataRae-AIEngineer you can pair them with local llms