“It has been 2 years of the Gemini program and GDM. We have now come a good distance in that point with many efforts we should always really feel very happy with. On the identical time competitors has accelerated immensely and the ultimate race to AGI is afoot. I believe we have now all of the elements to win this race however we’re going to need to turbocharge our efforts.
Code issues most — AGI will occur with takeoff, when the Al improves itself. In all probability initially it is going to be with a whole lot of human assist so a very powerful is our code efficiency. Moreover this must work on our personal 1p code. We have now to be probably the most environment friendly coder and Al scientists on the earth by utilizing our personal Al.
Productiveness — In my expertise about 60 hours every week is the candy spot of productiveness. Some people put in much more however can burn out or lose creativity. A lot of people work lower than 60 hours and a small quantity put within the naked minimal to get by. This final group shouldn’t be solely unproductive but additionally may be extremely demoralizing to everybody else.
Location — You will need to work within the workplace as a result of bodily being collectively is much simpler for communication than gve and so forth. And, due to this fact it is advisable to be bodily colocated with others engaged on the identical factor. We have to decrease reporting traces throughout international locations, cities, and buildings. I like to recommend being within the workplace not less than each week day.
Group — We have to have clear accountability and group with excessive functioning teams with shared administration and expertise management.
Simplicity — Lets use easy options the place we will. Eg if prompting works, simply do this, don’t posttrain a separate mannequin. No pointless technical complexities (similar to lora). Ideally we’ll actually have one recipe and one mannequin which might merely be prompted for various makes use of.
Excellence — whether or not it’s an eval or an information supply or a dashboard or a message in an inside Ul, please be certain that all of them work and all are good.
Velocity — we’d like our merchandise, fashions, inside instruments to be quick. Can’t wait 20 minutes to run a little bit of python on borg.
Iterate at small scale — we’d like numerous concepts that we will check rapidly. The easiest way to do that is small scale experiments till you may ramp up and hopefully see growing benefit at scale. This is a superb validation. Working an excessive amount of at simply massive scale has a behavior of minor tweaking and overfitting to evals, checkpoint sniping, and so forth. We’d like actual wins that scale.
No punting — we will’t hold constructing nanny merchandise. Our merchandise are overrun with filters and punts of varied varieties. We’d like succesful merchandise and [to] belief our customers.“