Where are the best researchers going?
While big tech (primarily FAANG) companies have historically attracted most top researchers out of PhD and post-doc programs, as venture capitalists get increasingly excited about productizing AI for consumers and businesses, we are seeing many of the best research scientists leave to start or join AI companies.
For example:
Mustafa Suleyman, the co-founder of DeepMind recently left to start Inflection AI alongside famed LinkedIn founder Reid Hoffman. After raising $225M+ at a $1.2B valuation ( 😯), they’ve released their first product called Pi - an AI companion and assistant. Given the status of the founders, they are apparently in talks to raise another $650M+.
This early in a startup’s lifecycle it’s unclear how well they will ultimately do. However, many early DeepMind researchers have left to join Inflection and bring millions of dollars worth of experience and expertise. If Inflective continues to gain traction - early employees will likely have opportunities to cash out early via a secondary sale.
And Inflective is just one of dozens of startups being started by prominent researchers.
Six of the eight original authors of the renowned Attention Is All You Need paper from Google have gone on to start billion-dollar companies and many of the best researchers have followed them:
- Noam Shazer started Character.AI and has raised $150M at a $1B valuation.
- Aidan Gomez started Cohere (an OpenAI model competitor). Cohere has raised hundreds of millions and is currently valued between $2-6B (latest funding round isn’t confirmed yet).
- Jakob Uszkoreit started Inceptive, an mRNA company using AI.
Life in big tech vs. a startup
So should you follow some of the best researchers and join a hot, quickly-growing AI startup?
It comes down to three things:
Type of work
When it comes to work, we see people choose to join or stay at big-tech research labs specifically because they are very research and publishing focused. If your goal is to publish as many papers as possible, it’s very likely the best place to be is big-tech.
Historically, big-tech has been willing to invest (via salaries and GPUs) in AI researchers to allow them to push the boundary of what is possible with the technology. However, the industry seems to be moving away from open source and publishing – and with the competition in AI increasing and interest rates rising – we will likely continue to see more closed-source research and more pressure to produce research that is commercially valuable today.
Work at most of the very hot AI startups includes a mix of research and product. It’s much rarer to be publishing papers but frequently researchers are doing cutting-edge research.
For example, one of Character.AI’s big focuses is on how to elegantly extend memory in their conversations. Their researchers are likely trying everything from extending context lengths in their LLMs to storing relevant past conversation notes in vectors and attempting to build query models over them. In many ways, the research at startups can be some of the most exciting because it’s immediately applied to a real-world company.
Financial Outlook / Risk
With numerous layoffs, big tech might not be as stable as it was a few years ago but the fact that their equity is public (i.e. you can sell it to anyone on the stock market) makes compensation at companies like Amazon, Google, NVIDIA, etc. more stable than joining an early-stage startup. (More on big tech research scientist compensation here)
Most of the well-funded AI startups will pay a base salary between $170-250K in the US. While more than enough to cover most living expenses in cities like NY or SF, it often doesn’t build life-changing savings. The equity you receive will likely either end up being worth zero or very, very valuable. It’s rare to have a middle scenario – where a startup has some sort of exit but early employees don’t see a significant financial outcome.
We recommend expecting no liquidity (i.e. ability to sell your equity) for 5-10 years when joining an early-stage company. We have former clients who joined companies like Cohere in the early days and are now worth 10s of millions of dollars. However, Cohere hasn’t allowed their employees to do a secondary offering (selling your shares to investors) so they haven’t been able to cash in yet.
It’s very hard to predict whether a company will succeed and - if they do - if they will allow secondary offers, so you need to be financially prepared for your equity to be worth $0 for a very long time (even if the company continues to grow and raise venture capital).
People & Pace
There are exceptional research teams in both big tech and startups. In the end, you spend the majority of your waking hours working and the people you work with can make a huge impact on your happiness and career satisfaction. You won’t know which culture or which team you gel with the most until you talk to them so we’d recommend considering both big-tech and startups with an open mind.
In terms of pace, startups tend to have a reputation for requiring longer hours and more fast-paced work – however, every company and team will be different. (Curious about what it’s like to work at a company? Here are some questions to ask the hiring manager and people on your team)
We are seeing an increasing number of the researchers we work with make the jump from big-tech to well-funded, incredibly exciting startups. It’s not the right choice for everyone, but it’s a path worth considering given how many benefits there are.
If you want our take on your specific situation you can book a free call here.
How to recruit at startups
While big tech recruiting tends to be fairly standardized (especially if you’re in a CS PhD program), recruiting at hot startups is often quite different. It’s much rarer to have recruiters reach out - normally you need to do the leg work.
Here’s how to get a job at a startup:
I’ve written a more detailed article about networking into startups including several of the AI companies I think are very compelling – and specifically which team member I’d reach out to on LinkedIn. You can read it here.
.png)
.png)
