Mo Bavarian

1.1K posts

Mo Bavarian

Mo Bavarian

@mobav0

Scaling up RL at OpenAI 🍓 Optimization and long context research before. Math PhD, MIT.

San Francisco, CA Katılım Ekim 2016
1K Takip Edilen17.7K Takipçiler
Mo Bavarian retweetledi
Design Arena
Design Arena@DesignArena·
BREAKING - OFFICIAL RESULTS: GPT-5.6 Sol by @OpenAI is 1st overall on Design Arena with an Elo of 1353. This puts GPT-5.6 Sol above Claude Fable 5 by @AnthropicAI and in the same performance band as GLM 5.2 by @Zai_org on frontend design. This is an 18-position and 60-point Elo leap from GPT-5.5. GPT-5.6 Sol also establishes a new Pareto frontier for preference vs. speed, faster than any model at this performance. Congratulations to the @OpenAI team on the launch!
Design Arena tweet media
English
111
190
1.9K
507K
Mo Bavarian
Mo Bavarian@mobav0·
From Tokenmaxxing to Healthmaxxing to Babymaxxing
English
1
0
16
2.1K
Mo Bavarian retweetledi
Jaynit
Jaynit@jaynitx·
Charlie Munger spent 50 years studying why intelligent people make catastrophically stupid decisions. It is the most useful thing I have ever watched: 1. Incentives are more powerful than anyone thinks. Munger says he has been in the top 5% of his age cohort his entire life in understanding the power of incentives and he has still underestimated it every single year. Federal Express could not get their night shift to work efficiently until someone realized they were paying by the hour. They switched to paying by the shift. The problem disappeared immediately. 2. People rationalise terrible behavior when their incentives point that way, and they do not even know they are doing it. A doctor in Nebraska was removing perfectly healthy gallbladders for years. When Munger asked an old colleague whether the doctor knew he was harming patients, the answer was no. he genuinely believed the gallbladder was the source of all medical evil and that removing it was an act of love. That is incentive-caused bias at its most extreme. 3. Psychological denial is real, and it is not just for weak people. A family friend's son flew off a carrier in the North Atlantic and never came back. His mother, a completely sane woman, simply never believed he was dead. Reality was too painful, so she distorted it until it was bearable. Munger says we all do this to some extent, and it causes terrible problems. 4. Consistency and commitment tendency are one of the most powerful forces in the human mind. Once you have stated a position publicly, you are psychologically locked into it. Max Planck said the really important new physics was never accepted by the old guard. A new guard came along that was less brain blocked by its previous conclusions. If this happened to the deans of physics, Munger says, imagine what it does to ordinary people. 5. The Chinese brainwashing system used on prisoners of war worked better than torture. They did not start with big demands. They maneuvered people into making tiny little commitments and declarations and slowly built from there. The same mechanism operates in every cult, every sales system, and every ideology that gets deeply embedded in people's heads. 6. Pavlovian association shapes buying behavior at a level most people never consciously process. Munger estimates three quarters of all advertising works on pure Pavlov. Coca-Cola does not want to be associated with funerals. They want to be associated with the Olympics, wonderful music, heroics. The association itself changes how people feel about the product at a subconscious level. Raising the price of a product can actually increase its market share because price and quality are associated in the human mind, and people use price as a signal of value. 7. Persian messenger syndrome is alive and running every major organization. The Persians killed the messenger who brought bad news. Bill Paley in his last 20 years, did not hear one thing he did not want to hear. everyone around him knew bringing bad news was dangerous. The result was that one of the most powerful men in media made terrible decisions for two decades because reality never reached him. 8. Social proof causes otherwise intelligent people to follow each other off cliffs. When one oil company bought a fertilizer company in the 1970s, practically every other major oil company rushed out and did the same. There was no rational reason for oil companies to own fertilizer companies. But if Exxon was doing it, it was good enough for Mobil. Every single acquisition was a disaster. 9. The efficient market theory persisted in academia for decades despite Berkshire Hathaway existing as a living contradiction. One economist kept adding sigmas to explain away the anomaly. two sigma, then three, then four, eventually six sigma. Munger's observation: It is better to add a sigma than change a theory just because the evidence comes in differently. That economist later went into money management himself and sank like a stone. 10. Contrast bias warps perception constantly and invisibly. Put your hand in hot water, then room temperature water. It feels cold. Put your hand in cold water, then room temperature water. It feels hot. same bucket. The human sensory apparatus has no absolute scale, only a contrast scale. Real estate agents exploit this deliberately. They show you two overpriced, awful houses first, then take you to a merely overpriced house, and it feels like a bargain. 11. The frog in slowly heating water is the business version of contrast bias. If something bad comes to you in small pieces, you are likely to miss it entirely. Munger says he has known many high-powered brilliant businessmen who were destroyed this way. not because they were stupid but because each incremental change was too small to trigger alarm. The contrast was never large enough to notice. 12. Authority bias is so powerful it can make trained professionals watch a plane crash. In flight simulator experiments, when the pilot, the authority figure, does something that any trained co-pilot knows will crash the plane, 25% of the time, the co-pilot sits there and lets it crash anyway. They have been trained to know better. The authority relationship overrides the training. 13. Deprivation super reaction syndrome explains why people go insane over small losses. Munger's neighbor had a 180 degree view of the harbor. the neighbor put in a pine tree about 3 feet high that turned it into a 179 and three-quarter degree view. They had a blood feud that went on for years. The New Coke disaster is the corporate version. Coca-Cola told customers they were changing a flavor and triggered a deprival super reaction so powerful that Pepsi was weeks away from releasing old Coke in a Pepsi bottle. smart engineers. brilliant lawyers. armies of psychologists. All missed it. 14. Envy and jealousy are far more powerful than greed and almost entirely absent from psychology textbooks. Munger says Warren Buffett has said half a dozen times that it is not greed that drives the world but envy. In a thousand-page psychology textbook, the index entry for envy and jealousy is blank. One of the most powerful forces in human behavior and academia essentially ignores it. 15. Gambling addiction is not explained by variable reinforcement alone. Skinner thought he had fully explained gambling by showing that variable reward schedules pound in behavior more powerfully than fixed ones. But the people who design modern slot machines know things Skinner did not. Lotteries where you pick your own number get far more play than lotteries where the number is assigned to you. People who commit to a number believe it has more validity because they chose it. Near misses on slot machines trigger deprival super reaction syndrome. It is four or five psychological tendencies working together, not one. 16. The most dangerous situations are when multiple psychological tendencies combine toward the same end at once. Munger calls this the lollapalooza effect. Tupperware parties use four or five tendencies simultaneously. Moonie conversion methods combine multiple tendencies and work extraordinarily well. alcoholics anonymous achieves a 50% no drinking rate when everything else fails because it also combines multiple tendencies toward a constructive end. The Milgram experiment is not just about obedience. it involves authority bias, consistency and commitment tendency, and contrast effects all working together. That combination turns human brains into mush. 17. Boards of directors are structurally designed to fail as corrective mechanisms. The top executive is the authority figure. He is doing something questionable. You look around, and nobody else is objecting, which is social proof that it is fine. He flies you around in the corporate jet and raises your director fees every year, which triggers reciprocation tendency. Munger's rule: boards only act when the behavior gets so bad it starts making them look foolish or threatens legal liability. That is the only forcing function that reliably works. 18. John Goodfriend of Salomon Brothers destroyed his career and reputation because he did not fire a trusted employee who had lied to the government. Every psychological tendency pointed toward keeping the man. He was a close colleague. His wife was known. He was part of a group that had made over a billion dollars for the firm. He said he had never done it before and would never do it again. Goodfriend looked into his eyes and believed him. The man did it again. The lesson: everyone who gets caught embezzling says they have never done it before and will never do it again. That is what they all say. 19. Darwin avoided confirmation bias by deliberately seeking out disconfirming evidence. Munger says Darwin was not especially smart by ordinary standards of human acuity. Yet he is buried in Westminster Abbey. Munger studied how Darwin worked and realized he had psychological tricks worth learning. Darwin always paid extra attention to evidence that contradicted his theories. Munger started doing the same and credits it as one of the most important intellectual habits of his life. 20. Why is the most important word in communication? Carl Braun designed oil refineries with spectacular skill, and you got fired in his company if you wrote a communication without explaining why. not just who, what, where, and when, but why. Braun knew that in a complex system where things can blow up, a communication system that always explains the reason behind an instruction works dramatically better than one that does not. Forstein, the general counsel of Salomon, told Goodfriend on multiple occasions that he had to report the employee's misconduct. He explained it was the right thing to do. He never explained what would happen to Goodfriend personally if he did not. he failed to use the most powerful tool of persuasion. Goodfriend ignored him. When Goodfriend went down, Forstein went with him.
English
58
798
3.2K
757.6K
Mo Bavarian
Mo Bavarian@mobav0·
@TheGregYang How are you measuring the energy score? Is it subjective or coming from any biomarkers?
English
1
0
10
943
Greg Yang
Greg Yang@TheGregYang·
my energy has improved gradually over time by mid May I could micromanage my life to feel normal (energy 7) but robustness is still not there: energy dipped after I loosened up after mid May and trying some new interventions nevertheless I feel things are improving under the hood! very grateful for all that I've learned on this journey so far 🙏
Greg Yang tweet media
English
15
1
237
17.6K
Mo Bavarian
Mo Bavarian@mobav0·
Honored to back this team!! 🚀
Behnam Neyshabur@bneyshabur

Today, I’m excited to formally announce @mirendil with my amazing co-founders Harsh Mehta, Shayan Salehian, and Tara Rezaei! We’re fortunate to work with @a16z and @kleinerperkins, who led our seed round of $200M, followed by a major investment from NVIDIA, among others. Mirendil exists to accelerate science and technology, and through them, to help solve humanity's most pressing problems. Self-accelerating AI R&D is the most direct path to delivering on AI's broader promise, which is why we believe the most important application of AI is AI itself. Get this loop right, and it compounds. It fundamentally changes the rate of progress itself across all domains. We believe this capability should be democratized. It should be used to power all scientific efforts trying to innovate at the frontier. There are far more important problems—and broader ones—than any single lab can take on, so more groups should be able to pursue them. This pulls concentration of power away from a few labs: businesses and science labs can own their AI and infrastructure, keep their margins, and control their own destiny instead of ceding it all to a single AI lab. We’re a small team with a singular focus. Our founding team consists of 20 researchers and engineers from frontier institutions including Anthropic, xAI, Google DeepMind, and OpenAI, united by a passion for science and a drive to build the technologies that move it faster. If you want to build the system that builds systems, join us! @HarshMeh1a, @shayan_, @tararezaeikh

English
2
1
25
5.3K
Noam Shazeer
Noam Shazeer@NoamShazeer·
I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there. It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to work with all of you.
English
988
868
16.3K
9.5M
Mo Bavarian
Mo Bavarian@mobav0·
In AI everything behaves according to a log-log scaling law, including the compensation
Mo Bavarian tweet media
English
7
1
65
7.1K
Mo Bavarian retweetledi
Dwarkesh Patel
Dwarkesh Patel@dwarkesh_sp·
New blackboard lecture w @reinerpope How do chips actually work – starting with basic logic gates, and working up to why GPUs, TPUs, FPGAs, and the human brain each look the way they do. 0:00:00 – Building a multiply-accumulate from logic gates 0:16:20 – Muxes and the cost of data movement 0:25:59 – How systolic arrays work 0:39:00 – Clock cycles and pipeline registers 0:51:40 – FPGAs vs ASICs 1:03:14 – Cache vs scratchpad 1:07:16 – Why CPU cores are much bigger than GPU cores 1:11:49 – Brains vs chips 1:15:22 – A GPU is just a bunch of tiny TPUs Look up Dwarkesh Podcast on YouTube/Spotify/etc to watch. Enjoy!
English
92
718
5.6K
933K
Mo Bavarian retweetledi
Timothy Gowers @wtgowers
Timothy Gowers @wtgowers@wtgowers·
AI has now solved a major open problem -- one of the best known Erdos problems called the unit distance problem, one of Erdos's favourite questions and one that many mathematicians had tried. openai.com/index/model-di…
English
75
613
3.6K
1.5M
Mo Bavarian retweetledi
Mo Bavarian
Mo Bavarian@mobav0·
Stop living in a bubble. The key to happiness, the true meaning of life, is not impact or wealth or fame, but health, friendship, and care for others. Living authentically is not that expensive. In other words, money or success will not fill the hole in your heart.
amadeo@amapel

x.com/i/article/2056…

English
3
4
103
34.8K
Mo Bavarian
Mo Bavarian@mobav0·
Stochastic gradient descent
Mo Bavarian tweet media
English
9
33
418
19.5K
Mo Bavarian
Mo Bavarian@mobav0·
@xanderai Tbc, I have no beef with peptides. May they all triumph in RCAs and become smashing successes like GLP1’s. But until the RCAs, the grifters are gonna grift
English
0
0
2
187
Xander Dunn
Xander Dunn@xanderai·
@mobav0 AI will exit its grifter phase when it makes a scientific discovery on the level of GLP1 peptides.
English
1
0
3
318
Mo Bavarian
Mo Bavarian@mobav0·
One of the biggest benefits of peptides is that it has distracted a decent fraction of grifters (ever-present in SF tech scene) away from AI. God bless 🙏
English
3
1
82
6.5K