Some great details about OpenAI's compensation here:
INNOVATION
-

OpenAI and Anthropic Staff Incentives Misaligned With Company Missions
By
–
Both OpenAI (through "Profit Participation Units") and Anthropic offer compensation that's primarily based on profitability. Yet OpenAI is run by a non-profit and Anthropic is a PBC. This means that staff incentives are by definition not aligned with the companies' missions.
-
Over-Engineering vs KISS: Premature Complexity in System Design
By
–
The opposite of KISS would be, for example: Assuming that we need to build a system for millions of users, with auto-scaling, automating migration of user data, end-to-end encryption … and starting to solve these problems on day one, before the first MVP of LTM was even built.
-
Unknown Researcher’s Breakthrough at OpenAI Challenges Academic Expectations
By
–
now seems as good a time as ever to remind people that the biggest breakthroughs at OpenAI came from a previously unknown researcher with a bachelors degree from olin college of engineering
-
Model Merging and Layer Count Expansion Research
By
–
Yeah there's plenty of lit on model merging in general, and IIRC it's always looking at fine-tunes from a common base — but I've not seen any papers about increasing layer counts like this.
-
Random Discovery Behind Successful AI Model Upsizing Technique
By
–
It's wild isn't it? I did a bit of a dive into this a few months ago and asked the guy who seemed to have done the first successful model upsizing where the idea came from. He told me it was just a random thing he tried and he had no particular reason to believe it would work.
-
Over-Engineering Small Projects: Recognizing Unnecessary Complexity
By
–
Yes, many things are "it depends." But you can spot the pattern I describe when the project is still small, but the solutions are already overly complicated.
-
Rapid Prototyping Over Perfect Code in AI Development
By
–
I should add: Absence of rapid prototyping. Engaging in excessively long iteration loops. For example, spending a month writing 'perfect code' and only then seeing if it was a good idea in the first place. In contrast, a high-level prototype developed in three days could
-
Scaling Alone Won’t Achieve AGI Says Researcher
By
–
this is a very good point, I guess this tweet comes with the association that I really don’t think scaling will get us to AGI
-
GPT Development Credits: Recognizing Key Contributors Beyond Papers
By
–
why! He’s a great communicator and thinker, I really admire him! But he wasn’t there when they built GPT-{1,2,3}, he isn’t on any of the papers