đ»Â Career advice for software engineers interested in Generative AI
đžÂ Including advice from Sam Altman and Marc Andreessen
đ Plus a list of the best generative ai startups to join
đŠŸÂ No machine learning background required
â Marc Andreessen
âThrough the Humanloop platform we see hundreds of AI product companies, and this is the guide that I'll point engineers interested in AI to.ââ Raza Habib, CEO of Humanloop.
Humanloop is a platform for making production GPT-3 apps and has raised from YC and Index Ventures.
đ° Premise
- Yes, AI will impact software engineering jobs. How could it not? Some devs are saving hours per day.
- Jobs will be destroyed. Jobs will also be created.
- Should devs be afraid of AI? Yes. Be afraid. But also be excited. And more important than what you feel is what your feelings are telling you to do! Fear is telling us to run away. But we canât run away from this. So we need to run toward it.
- When weâre scared of something, often the best thing to do is to make friends with it. This is true with our own fears inside our head, and itâs true of AI. đ€
- Generative AI will be massive. Anyone who says itâs like crypto is wrong. đ
- Crypto still hasnât proven superior for any paid use cases other than speculation â which, granted, is still a nice business (Coinbase, OpenSea), gambling (NFTs â which if you look up close at the reality of it is quite a sad story â in trading, for one party to win, another party has to lose â and it really ended up being a bunch of regular people losing a bunch of money to glorified casino operators), storage of value (BTC, ETH), and cross border payments (BTC, ETH, USDC, USDT, SOL). There are also crypto infrastructure startups that generate revenue (Chainalysis, Alchemy), but those are incestuous startups that only exist because of crypto. Kind of like how lawyers create complex laws that make lawyers more necessaryâŠ
- Not being involved with generative AI is a mistake. Itâs the fastest moving industry in the world right now. Get close and stay close! Work in gen ai, start a gen ai startup, build a gen ai side project, or at least use the generative ai tools as an end user, so you can see how itâs changing and evolving.
- Lots of people are scared of AI. Wondering if theyâll be automated. The best protection is to ride the wave.
đ Gen AI is a hypergrowth / rocketship / breakout industry.
đ For ambitious young people
- Building generative ai side projects is one of the best things you can do
- Then go and work at a generative ai startup, ideally one that has product-market fit (for example, theyâre doing $1+ million per year in revenue from their gen ai products). Or one that doesnât yet have product market fit but where youâre obsessed with the product and/or the team (you like them, you trust them, you respect them)
Ironically, this is an old quote from Sam Altman about Microsoft. Sam Altman negged Microsoft so hard that they ended up wanting to give him $10b đ
âIf you are young and want to have an impact, you want to be in an industry where there is a lot of growth and change and flux and opportunity. Once you have picked an industry, get right to the center of it as fast as you possibly can. Optimize at all times for being in the most dynamic and exciting pond you can find.â â Marc Andreessen
â ïž Warning: Donât just look for AI/ML companies. Know the difference between old ai and new ai
Old AI has been around for a while, is nice to have, but isnât the fast growing industry like generative ai is.
đ Old ai vs new ai
- đŸ Old AI
- Machine Learning / Data Science
- Taking in numbers and predicting the next numbers
- Examples:
- Most of the AI infrastructure companies are infrastructure for old AI. If you want to be in a hyper growth industry, avoid these
- Anything that says they use data to predict something is probably old AI
- AI for fraud detection, AI for manufacturing optimization, AI powered lending,
- If they were founded before 2022, thereâs a pretty good chance itâs old AI
- đ New AI
- If itâs doing something with software that requires a near-human-level understanding of text, then itâs new ai
- If itâs generating near-human-level text, then itâs new ai
- If itâs generating AI powered images, then itâs probably new ai
- If it uses GPT-3 or stable diffusion, then itâs new ai
- Some companies are a blend of old ai and new ai. For example scale ai is an infrastructure company, theyâre mostly built around old ai, but they also have some new ai related infrastructure
- Is the company generating images or text, or other content? If so, then it might be new ai
Old AI = less interesting than generative ai. And if youâre wondering âhow can it be true that ai is a good place to work when thereâs a bunch of ML-related layoffsâ â part of the answer is that lots of companies invested in ML/data science experimental projects that werenât core to the success of their company. So now that the economy is doing worse, those areas are being cut.
đ Look for companies with new ai at the core, not at the edges
- đ„ Good
- a company that has a core product that provides apis or infrastructure tools for generative ai companies (and not for old ai companies)
- eg openai
- a company where the company couldnât exist without generative AI
- eg jasper, character ai
- a team working on a product that couldnât exist without generative AI, where the product could be big enough to be its own business, but where the overall company doesnât rely on generative ai
- eg being at github working on copilot
- đ„ Decent
- A key feature that relies on generative ai, but the feature wouldnât be its own business, and where the company and product would still be fine if generative ai didnât exist
- eg notion ai
Thereâs also a separate tier ranking thatâs something like:
- making the general models (eg openai)
- making specific models (eg ?)
- making complex gen ai products (eg runwayml)
- making simpler api wrapper gen ai products (eg jasper)
- making prompts
đ GPT vs AI vs ML
Roughly itâs like this:
- AI
- Old AI
- Old ML
- New AI
- New ML
- Text generation
- OpenAI
- GPT-3
- ChatGPT
- Image generation
- Stability AI
- OpenAI
- Dall-E
đ The gen ai stack
- End user company (e.g. Jasper)
- API (eg OpenAI)
- Model (eg GPT-3)
- Training data
- Reinforcement learning from human feedback
- Compute power (eg A100s)
List of the best generative ai startups and large language model startups
Large companies
- Microsoft (if they do more with OpenAI, they might end up as number one)
- Nvidia (A100sâŠ)
- Meta
- Apple
Though note that smaller companies are a better fit for doing interesting things with generative ai, at least for now. So unless youâd only feel comfortable working at a large company, itâll be better to go to a smaller one.
Medium-large
- OpenAI
- Deepmind (though Deepmind is more oriented around reinforcement learning rather than language models and deep learning, and they are more focused on research than building products)
- Stability AI
Medium sized companies that havenât released a public product yetâŠ
â ïžÂ Warning: The longer a company goes without releasing a product, the more likely it is that theyâll never release a product
- Anthropic (product coming soon â Claude)
- Inflection
Tiers for model companies
- OpenAI
- Anthropic
- The rest
Consumer product startups
- Jasper
- Runway ML
đ§ What skills and roles are needed?
- Product-oriented engineers. Now that we have great models available through APIs, we need product-oriented engineers who can build new things with them, and machine learning backgrounds are less important.
âïžÂ Next steps
- Join some generative AI discords. CarperAI, OpenAI, EleutherAI, Stable Diffusion.
- Build a gen ai product.
- Apply to a gen ai job.