Sergey Brin urges Google’s AI developers to stop coddling users and work harder

With the AI ​​boom underway, Google co-founder Sergey Brin, who had been taking a well-deserved rest, has returned to full-time work at the company. This week, he addressed his subordinates, reminding them that competition has intensified and giving advice on how to continue working in the team, The Verge reports.

Sergey Brin. Image source: wikipedia.org

The biggest players are in the final stages of a race to become the first to develop strong AI (AGI), Brin said. To achieve this, Google developers will have to work faster, prioritize simple solutions, stop wasting time, and stop coddling users who need to be trusted. The memo is addressed to employees of Google’s DeepMind division, but it provides a vivid insight into the AI ​​industry. Here is the full text:

«The Gemini and GDM (Google DeepMind) program is two years old. We have come a long way and put in a lot of effort that we should be proud of. At the same time, the competition has intensified tremendously and the final race for strong AI has begun. I think we have all the ingredients to win this race, but it will take everything we have.

Code is king – strong AI will take off when AI starts improving itself. It will likely require significant human assistance at first, so code performance is king. And that includes code we write. We need to be the most efficient programmers and AI researchers in the world, using our own AI.

Productivity – In my experience, 60 hours a week is the sweet spot for work tempo. Some people work much more, but they can burn out or lose their creativity. Some work less than 60 hours, but then there are those who work the bare minimum, just to get by. These people are not only ineffective, but can seriously undermine the discipline of the entire team.

Workplace – It is important to work in an office because personal presence is much more effective than virtual communication. We need to be physically close to colleagues working on the same tasks. We need to minimize the number of subordination levels between countries, cities and office buildings. I recommend being in the office at least every weekday.

Organization – We need clear lines of accountability and a structure with highly effective departments, overall management and technology leadership.

Simplicity – Let’s use simple solutions where possible. For example, if queries work, let them just work – no need to train a separate model further. No unnecessary technical complexity (like LoRA). Ideally, we should have one architecture and one model that can be queried for different purposes.

Quality – Whether it’s a score, a data source, a dashboard, or a backend message, make sure it works and is executed to a high standard.

Speed ​​- Our products, models, and internal tools need to be fast. I can’t wait 20 minutes for Borg to run a small piece of Python code.

Small scale iterations – we need lots of ideas that can be tested quickly. The best way to do this is to experiment on a small scale and then, if successful, scale them up. This is a great way to validate. When you go big right away, there is a tendency to tweak things, adjust to metrics, chase milestones, etc. At scale, we need real wins.

Stop passing the buck — we can’t build nanny products. Our products are full of filters and compromises. We need solutions that work, and we need to trust our users.”