Category 2: AI Labs -- Foundational Model Research
Working at a major AI lab builds deep expertise in foundational model research -- the exact skills needed to eventually build an AI Warren Buffett. These labs are at the frontier of capabilities research, alignment, and scaling. All 6 labs have significant Bay Area presence.
Bay Area Companies (6)
Click to jump to a company
6
Bay Area Companies
0
Researching
0
Interested
0
Applied
1
Anthropic Bay Area
Constitutional AI, mechanistic interpretability, and the Claude model family
$61.5B+
Valuation
1,500+
Employees
2021
Founded
San Francisco
HQ
Private
Stage
Claude
Model Family
Key People
Dario Amodei
CEO & Co-founder
Former VP of Research at OpenAI. Princeton PhD in computational neuroscience. Led GPT-2 and GPT-3 safety work before founding Anthropic with a focus on AI safety research.
Daniela Amodei
President & Co-founder
Former VP of Operations at OpenAI. Oversees business operations, policy, and go-to-market strategy at Anthropic.
Chris Olah
Interpretability Lead
Pioneer of neural network interpretability. Created distill.pub. Leading mechanistic interpretability research -- understanding what happens inside neural networks at the circuit level.
Tom Brown
Co-founder
Lead author of the GPT-3 paper ("Language Models are Few-Shot Learners"). One of the most impactful ML papers ever written.
Sam McCandlish
Co-founder
Co-author of the neural scaling laws papers at OpenAI. Research on how model performance scales with compute, data, and parameters.
Jared Kaplan
Co-founder
Johns Hopkins physicist turned ML researcher. Lead author of the original neural scaling laws paper that shaped how the entire field thinks about model training.
Product: Claude Models & Constitutional AI
- Claude model family (Haiku, Sonnet, Opus) -- ranging from fast/cheap to highly capable frontier models
- Constitutional AI (CAI) -- training LLMs using a set of principles rather than pure human feedback, enabling RLHF/RLAIF training methodology
- Mechanistic interpretability -- understanding what happens inside neural networks at the circuit level, identifying specific neurons and circuits responsible for specific behaviors
- Safety evaluations and red-teaming -- systematic approaches to finding and mitigating model risks before deployment
- Claude API -- developer platform for building on Claude, including function calling, tool use, and computer use capabilities
- Claude.ai -- consumer product with artifacts, projects, and extended thinking capabilities
Compensation (Research Engineer / ML Engineer)
- Research Engineer L5: $350K-$450K total comp
- Senior / Staff: $500K-$800K+
- Significant equity -- pre-IPO at $61.5B valuation
- Source: levels.fyi
Why It Matters
Safety-first research lab building frontier models. Constitutional AI and mechanistic interpretability are foundational research areas -- understanding how models work and how to control them is essential for building trustworthy AI systems. The interpretability work (Chris Olah's team) provides unique insight into what neural networks actually learn, which is directly relevant to building an AI that can reason about investment opportunities. Pre-IPO equity at $61.5B represents significant upside. SF location is ideal.
Path to Entry
- Research Engineer (best fit for ML engineers) -- strong coding + ML understanding
- Research Scientist roles in Alignment, Interpretability, Multimodal
- 5-6 round interview process
- Referrals matter significantly -- networking is critical
- Meta ML experience valued for production-scale systems knowledge
Tracking
2
OpenAI Bay Area
GPT-4, o1/o3 reasoning models, and ChatGPT -- market leader in commercial LLM deployment
$150B+
Valuation
3,000+
Employees
2015
Founded
San Francisco
HQ
Private
Stage
GPT-4 / o3
Frontier Models
Key People
Sam Altman
CEO
Former president of Y Combinator. Drove OpenAI's transformation from research nonprofit to the most commercially successful AI company. Navigated the 2023 board crisis and emerged stronger.
Mark Chen
CTO
Formerly led DALL-E and Codex development. Promoted to CTO as OpenAI scales its research and engineering organization.
Jakub Pachocki
Chief Scientist
Successor to Ilya Sutskever as Chief Scientist. Leads OpenAI's core research direction including scaling, alignment, and next-generation model architectures.
Product: GPT-4, o1/o3, ChatGPT, and Platform
- GPT-4 and GPT-4o -- frontier multimodal models handling text, images, audio, and video in a single model
- o1/o3 reasoning models -- chain-of-thought reasoning models that "think" before answering, representing a new paradigm in AI capabilities beyond raw pattern matching
- ChatGPT -- fastest-growing consumer product in history, 100M+ weekly active users
- DALL-E -- text-to-image generation model
- Whisper -- open-source speech-to-text model, state-of-the-art accuracy
- Sora -- video generation model producing realistic video from text prompts
- API platform -- serving millions of developers with function calling, tool use, fine-tuning, and embeddings
Compensation (Research Engineer / Applied AI)
- Research Engineer: $400K-$550K total comp
- Senior: $600K-$900K+
- Profit Participation Units (PPUs) -- among highest-paying AI companies
- Source: levels.fyi, Blind
Why It Matters
Market leader in commercial LLM deployment with unmatched scale in production AI. The o1/o3 reasoning models represent a fundamentally new paradigm -- teaching models to "think" step-by-step rather than just pattern match. This reasoning capability is exactly what an AI Warren Buffett would need: the ability to reason through complex investment scenarios. OpenAI also has the most data on how LLMs behave at scale, which is invaluable experience. Highest compensation in the industry. SF location is ideal.
Path to Entry
- Research Engineer, Applied AI Researcher, Platform Engineer roles
- Take-home project + 4-5 onsite rounds
- Strong preference for production ML at scale -- Meta experience is highly valued
- Emphasis on systems thinking and ability to ship at scale
- Competitive process -- strong referral network helps significantly
Tracking
3
Google DeepMind Bay Area
Gemini models, AlphaFold (Nobel Prize), and the most research-productive AI lab in history
Alphabet
Parent ($2T+)
2,500+
Employees
2010
Founded
MTV / London
HQ
2023
Brain Merger
Gemini
Model Family
Key People
Demis Hassabis
CEO
Nobel Prize in Chemistry 2024 for AlphaFold. Co-founded DeepMind in 2010. Former chess prodigy and game designer. Led the creation of AlphaGo, AlphaFold, and Gemini.
Jeff Dean
Chief Scientist
Creator of MapReduce, BigTable, and TensorFlow. One of the most impactful engineers in computing history. Now leads Google's AI research direction across DeepMind.
Shane Legg
Co-founder & Chief AGI Scientist
Co-founded DeepMind with Hassabis. His PhD thesis on universal intelligence formalized AGI measurement. Leads long-term AGI safety and research direction.
Product: Gemini, AlphaFold, and Research Portfolio
- Gemini models (Ultra, Pro, Flash, Nano) -- multimodal models with 1M+ token context windows, competing at the frontier with GPT-4 and Claude
- AlphaFold -- solved protein structure prediction, won the 2024 Nobel Prize in Chemistry, one of the most impactful scientific contributions from AI
- AlphaGo/AlphaZero -- defeated world Go champion Lee Sedol, demonstrated superhuman performance through self-play reinforcement learning
- AlphaProof -- mathematical reasoning system achieving silver-medal performance at the International Mathematical Olympiad
- GraphCast -- weather prediction model outperforming traditional numerical weather models
- RT-2 -- robotic transformer model connecting language understanding to physical robot control
- Google AI Studio and Gemini API -- developer platform for building on Gemini models
Compensation (Software Engineer / Research)
- L5 Software Engineer: $350K-$450K total comp
- L6 Senior: $500K-$700K
- L7 Staff: $800K+
- Google RSUs, excellent benefits, unlimited TPU access
- Source: levels.fyi
Why It Matters
Most research-productive AI lab in history. Access to Google's unlimited TPU compute and academic research culture with massive resources. AlphaFold proved that AI can solve fundamental scientific problems -- the same approach could be applied to financial analysis. AlphaZero's self-play methodology is directly relevant to training an AI to explore investment strategies. The breadth of research (language, vision, science, math, robotics) provides unmatched cross-domain learning. Mountain View office is Bay Area.
Path to Entry
- Standard Google interview + ML-specific rounds
- Research Scientist (PhD preferred), Research Engineer, SWE (ML) roles
- DeepMind-specific roles improve team placement within the merged org
- Mountain View office is the primary Bay Area location
- Google's structured interview process is well-documented -- prepare for coding + ML design + research depth
Tracking
4
Meta FAIR Bay Area
Open-source LLaMA family, Yann LeCun's world models, and PyTorch framework
Meta
Parent ($1.5T+)
500+
Researchers
2013
Founded
Menlo Park
HQ
Public (META)
Stage
LLaMA
Model Family
Key People
Yann LeCun
Chief AI Scientist
Turing Award winner (2018). Pioneer of convolutional neural networks (CNNs). Advocates strongly for self-supervised learning and JEPA world models as the path to human-level AI, rather than autoregressive language models.
Robert Fergus
Head of FAIR (as of 2025)
Formerly at Google DeepMind. NYU professor. Expert in computer vision and deep learning. Now leads FAIR's research direction.
Product: LLaMA, JEPA, and Open-Source AI
- LLaMA open-source LLM family -- democratized access to powerful language models, enabling the entire open-source AI ecosystem
- Self-supervised learning research (JEPA world models) -- LeCun's vision for AI that learns world models through prediction, not just next-token generation
- DINOv2 -- self-supervised vision model producing universal visual features without labels
- SAM (Segment Anything) -- foundation model for image segmentation that can segment any object in any image
- SeamlessM4T -- multilingual multimodal speech and text translation across 100+ languages
- Code Llama -- code-specialized LLM for code generation, completion, and understanding
- ESMFold -- protein structure prediction rivaling AlphaFold using language model techniques
- PyTorch -- the dominant deep learning framework used by most researchers worldwide
Compensation (ML Engineer / Research Scientist)
- E5 ML Engineer: $350K-$450K total comp
- E6 Senior: $500K-$700K
- E7 Staff: $800K+
- Meta RSUs, excellent benefits
- Source: levels.fyi
Why It Matters
Strongest open-source AI lab in the world. LeCun's contrarian views on world models vs autoregressive approaches could represent the next paradigm shift in AI. FAIR provides unlimited compute and access to Instagram/Facebook data at unprecedented scale. Most importantly for Ravi: internal transfer path is available since he is already at Meta. This is the lowest-friction option with the highest research quality. Menlo Park HQ is ideal for Bay Area.
Path to Entry
- INTERNAL TRANSFER (recommended for Ravi) -- reach out to FAIR team leads directly
- Internal interviews are lighter (2-3 rounds vs full external loop)
- Strategy: start with research collaboration on a project, then pursue full transfer
- External path: standard Meta interview (coding + ML system design + behavioral)
- Ravi's existing Meta tenure and ML experience make this the most accessible frontier lab
Tracking
$50B+
Valuation
100-200+
Employees
2023
Founded
Bay Area
HQ
Private
Stage
Grok
Model Family
Key People
Elon Musk
Founder & CEO
CEO of Tesla and SpaceX, owner of X (Twitter). Founded xAI to build AI that seeks truth and understanding. Known for extreme intensity and velocity in execution.
Igor Babuschkin
Co-founder
Former Google DeepMind researcher. Expert in large-scale model training and distributed systems for AI. Key technical leader driving xAI's rapid model development.
Product: Grok Models & Colossus Infrastructure
- Grok models (1, 2, 3, 4+) -- LLM integrated into X (Twitter) with real-time access to platform data
- Colossus cluster -- 100,000+ NVIDIA H100 GPUs, one of the world's largest AI training clusters, built in record time
- Real-time information access -- unique data advantage via X platform's live feed of global conversation and news
- "Truth-seeking" AI -- positioned as less restricted than competitors, focused on maximum helpfulness and accuracy
- Custom hardware optimization -- deep systems engineering for massive-scale training efficiency across the Colossus cluster
- Rapid iteration -- Grok 1 to Grok 4+ in under 2 years, demonstrating extreme development velocity
Compensation (ML Engineer / Research Engineer)
- Early-stage startup with high equity upside
- Estimated $400K-$700K total comp for ML Engineers
- Significant equity at $50B+ valuation
- Source: Blind, estimated (limited public data)
Why It Matters
Massive compute advantage with the Colossus cluster (100K+ H100s). Small team means enormous individual impact -- each engineer has outsized influence on model development. Musk's intensity drives extreme velocity, and the X platform data provides a unique real-time information advantage no other lab has. The small team size also means exposure to the full stack from infrastructure to model architecture to deployment. Bay Area based.
Path to Entry
- ML Engineer, Research Engineer, Infrastructure Engineer roles
- Smaller, less structured interview process compared to big labs
- Emphasis on practical ML skills and ability to ship fast
- Systems-level knowledge is critical -- GPU clusters, distributed training, model optimization
- Meta's production ML at scale experience is directly relevant
Tracking
6
Apple AI/ML Bay Area
Apple Intelligence, on-device LLMs, Private Cloud Compute, and MLX framework
$3T+
Market Cap
160K+
Total Employees
Cupertino, CA
HQ
Public (AAPL)
Stage
2B+
Active Devices
Apple Intelligence
AI Platform
Key People
John Giannandrea
SVP of ML & AI Strategy
Former head of Google Search and AI. Recruited by Tim Cook in 2018 to lead Apple's AI transformation. Oversees all ML/AI strategy across Apple products.
Craig Federighi
SVP of Software Engineering
Leads iOS, macOS, and platform software development. Key stakeholder in integrating Apple Intelligence across all Apple operating systems and devices.
Product: Apple Intelligence & On-Device ML
- Apple Intelligence -- on-device LLMs running directly on iPhone/iPad/Mac using Apple Neural Engine, with Private Cloud Compute for heavier tasks
- Private Cloud Compute -- server-side inference with cryptographic guarantees that Apple cannot see user data, a unique privacy-preserving approach
- CoreML framework -- developer toolkit for deploying ML models on Apple devices with hardware-accelerated inference
- Apple Neural Engine -- custom silicon (in M-series and A-series chips) optimized specifically for ML inference, enabling on-device AI
- MLX framework -- open-source ML framework optimized for Apple Silicon, enabling efficient model training and inference on Mac
- Siri (LLM-powered) -- personal assistant being rebuilt on foundation model technology
- Model compression and quantization -- state-of-the-art techniques for running billion-parameter models on mobile devices
- Privacy-preserving ML -- federated learning, differential privacy, on-device processing as core design principles
Compensation (ML Engineer / AI Research)
- ICT4 ML Engineer: $300K-$400K total comp
- ICT5 Senior: $450K-$600K
- ICT6 Staff: $700K+
- Apple RSUs, excellent benefits
- Source: levels.fyi
Why It Matters
Privacy-first AI approach is unique among all major labs. Unmatched hardware-software integration means Apple controls the full stack from custom silicon to model deployment. 2B+ active devices provide unmatched deployment reach for on-device AI. The model compression and quantization expertise (running LLMs on phones) is deeply technical and transferable. Cupertino HQ is ideal for Bay Area. While not a pure research lab, the scale of deployment and focus on efficient inference are unmatched.
Path to Entry
- ML Engineer, Research Scientist, AI/ML Engineer (Cloud) roles
- Apple interview: 5-6 rounds, heavy on past project deep dives
- Team match happens at interview time -- important to target the right AI/ML team
- Apple's secrecy culture means less public information about internal research
- Meta experience in production ML and model optimization is relevant