- Homescreen
- Posts
- š What Did Ilya See?
š What Did Ilya See?
OpenAI's secret breakthrough
Gm. Happy belated Thanksgiving, and happy Black Friday! This year, Iām grateful for a few things. My health, my girlfriend, Aldi, and, of course, all of you for continuing to tune in newsletter after newsletter. Without you, I wouldnāt have a jobā¦literally.
ā Stephen Flanders

š AI
OpenAIās Q*

I intentionally avoided an OpenAI lead story on Wednesday, but much like Sam Altman, we are so back.
This story is just too good to pass up.
The big mystery surrounding Samas firing was always āwhy?ā. Why fire the face of your company with no warning right before youāre about to raise a boatload of money? Unfortunately, the board decided to go zero-dark thirty and keep their reasoning private.
However, it appears we now have some answers.
According to reports, OpenAI had developed a model known as Q* (Q-Star) which was able to solve basic math problems it had not seen before. This was something that models were not able to do before because math isnāt like writing. You canāt just predict the next word. You have to be able to reason your way to the answer.
So, Q* is a model with at least some elementary ability to reason. Which would add some context to comments Sama made recently:
if Q* is an agentic algorithmic breakthrough that can reason and goal seek via something like A* search against an entire world model embedding, it completely explains this weird statement from sam last week:
"Is this a tool we've built or a creature?"
twitter.com/i/web/status/1ā¦
ā Siqi Chen (@blader)
1:20 AM ⢠Nov 23, 2023
Couple that reasoning breakthrough with OpenAIās ability to predict how intelligent a model will be based on compute, and you can at least begin to form a roadmap for AGI.
If youāre a little bit freaked right now, youāre not alone. Some of OpenAIās researchers were as well, so they sent a letter to the board which pretty much said āholy crapā, and next thing you know Sama is out of a job.
Is this really the breakthrough that leads to AGI? I have no idea, but Iām pretty confident with Sama back at the helm and more powerful than ever, weāll find out sooner or later.
Survey Says: Betting markets have the date for AGI as February 10th, 2031. Thatāll be an interesting Valentineās Day.

š¤ THE LATEST INā¦
TECH
Tesla has open-sourced the original roadster.
Rejoice, headlines are back on Tweets.
Reels are now downloadable.
Donāt fall for these chrome update hacks.
AI
Training your own AI model isnāt as hard as you think.
Bard is getting smarter and now understands YouTube videos.
Inflection-2 is here, and itās claimed to be the second-best model in the world (after GPT-4).
Say hello to the killer robots.
SCIENCE
Blue Origin is going to the moon.
Zoom fatigue is very much a real thing.
The 200-year-old Dolomite Problem has been solved.
CRYPTO
Coinbase is already taking advantage of the Binance situation.
South Korea is piloting a CBDC.
SBF is now paying for his haircuts with fish.
If you find yourself needing to crypto the coins over the holidays, show them this.

šāāļø QUICKIES
Raise: Aether Biomedical, a bionic hand startup, raised $5.8M.
Stat: 51.6%: The amount of tech workers that think AI is just hype.
Rabbit hole: Why I Quit Full-Time Roles (Elena Verna)

š ļø FOUNDERS CORNER
The best resources we recently came across that will help you become a better founder, builder, or investor.
š Raffle Leader will get you leads faster than the speed of light.
š¤ If you want to get better at prompt engineering, check this out.
š° Bet on deep tech, you wonāt regret it.

š FINAL JEOPARDY!
This Final Jeopardy! question stumped every contestant. Can you get it right?
Category: Television
Clue: This series grew out of a screenplay titled āMurdochā

š DOPAMINE HIT
How do you do this move?

HOW WAS TODAY'S NEWSLETTER? |

REACH 40K+ FOUNDERS, INVESTORS & OPERATORS
If youāre interested in advertising with us, send an email over to [email protected] with the subject āHomescreen Adsā.

FINAL JEOPARDY! ANSWER
Succession