Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Two years have passed since the Gemini and GDM program. We have come a long way at the time with many efforts that we should feel proud. At the same time, the competition accelerated greatly and the final race to AGI is in full swing. I think we have all the ingredients to win this race, but we have to have to charge our efforts.
Code Matters more – AGI will happen with takeoff, when Al himself improves. Perhaps at first it will be with a lot of human help, so the most important thing is our code performance. Moreover, this needs to work on our 1P icon. We must be the programmer and the most efficient scientists in the world using Al.
Productivity – In my experience about 60 hours per week is the beautiful place for productivity. Some people put much more than that, but they can burn or lose creativity. A number of people work less than 60 hours and a small number is minimal to get. This last group is not only fruitful, but can be very frustrated for anyone else.
The site – it is important to work in the office because being together is more effective to call GVE and so on, and thus you need to be physically with others who work on the same thing. We need to reduce reporting lines across countries, cities and buildings. I recommend that I be in the office at least every week.
Organization – We need a clear responsibility and organize with high -performance groups with the leadership of management and joint technology.
Simplicity – let’s use simple solutions where we can. For example, if that works to work, you just have to do this, do not publish a separate model. No unnecessary technical complications (such as Lora). Ideally, we will have one recipe and one model that can be simply required for different uses.
Excellence – whether it is an evaluation, a data source, a dashboard, or a message in an internal UL, please make sure it is all working and all good.
Speed - we need our products, models and internal tools to be fast. You cannot wait 20 minutes to run a little bit of Beton on Burg.
Repeating on a small scale – we need a lot of ideas that we can quickly test. The best way to do this is small -sized experiences so that you can increase and we hope to see an increasingly increasing feature. This is excellent verification. Working is very widely only has a slight change and outperforms on EVALS, checkpoint sniping, etc. We need real victories with this scale.
Do not swing – we cannot continue to build educator products. Our products with filters and lead are exceeded from different types. We need capable products and (to trust our users. “