seq2seq forecasting, time series models
Sequence-to-sequence models with attention forecast multiple time steps using encoder-decoder architectures.
311 technical terms and definitions
Sequence-to-sequence models with attention forecast multiple time steps using encoder-decoder architectures.
Sequence bias adjusts probabilities for multi-token patterns.
Split sequence dimension across GPUs to handle long contexts.
Sequential Monte Carlo methods use particle filtering for Bayesian inference in nonlinear state space models.
Server-sent events provide protocol for streaming responses from server to client.
Service level quantifies inventory availability as percentage of demand satisfied from stock.
Permutation-invariant transformer.
Set2Set uses attention mechanisms to create permutation-invariant graph-level representations from node features.
# San Francisco
# The Global Capital of Artificial Intelligence
## 1. Center of Artificial Intelligence Revolution
San Francisco has emerged as the **undisputed epicenter** of the artificial intelligence revolution. The city hosts:
- **83+ AI companies** leasing nearly 1 million sq ft of office space
- **78.2%** of all US venture capital funding flowing to Bay Area companies
- Projected **50,000+ AI workers** by 2030
- AI office space expected to grow from **5M to 21M sq ft** by 2030
## 2. AI Titans — Frontier AI Labs
### 2.1 OpenAI
- **Headquarters:** 3180 18th Street, San Francisco (Mission District)
- **Expanded Campus:** 550 Terry A. Francois Blvd, Mission Bay (315,000 sq ft)
- **Additional Space:** 1455 & 1515 Third St (486,600 sq ft, subleased from Uber)
- **Total Raised:** $65.1B
- **Key Products:** GPT-4, ChatGPT, DALL-E, Sora
**Leadership:**
- Sam Altman (CEO)
- Greg Brockman (President)
- Founding team included: Ilya Sutskever, Andrej Karpathy, Wojciech Zaremba
### 2.2 Anthropic
- **Headquarters:** 548 Market Street, PMB 90375, San Francisco, CA 94104, USA.
- **Key Products:** Claude AI (Opus, Sonnet, Haiku)
**Leadership:**
- Dario Amodei (CEO & Co-founder)
- Daniela Amodei (President & Co-founder)
- Founded: 2021 by former OpenAI researchers
**Anthropic's AI Alley Location:**
```
-
┌─────────────────────────────────────┐
│ Salesforce Tower │
│ ↓ │
│ [300 Howard St] ← Planned HQ │
│ ↓ │
│ [500 Howard St] ← Current HQ │
│ ↓ │
│ [505 Howard St] ← Expansion │
└─────────────────────────────────────┘
```
### 2.3 xAI (Elon Musk)
- **SF Office:** The Pioneer Building, 3180 18th St, San Francisco, CA.
- **Memphis Data Center:** 3231 Riverport Rd, Memphis, TN 38109 (Colossus 1).
- **Palo Alto Headquarters:** 1450 Page Mill Road, Palo Alto, CA 94304, USA (main research/leadership).
- **Key Products:** Grok AI
### 2.4 Thinking Machines Lab
- **Headquarters:** San Francisco
- **Founded:** February 2025
- **Structure:** Public Benefit Corporation
- **Funding:** $2B seed round (July 2025) at $12B valuation
- **Lead Investor:** Andreessen Horowitz (a16z)
**Founding Team:**
- Mira Murati (CEO) — Former OpenAI CTO
- John Schulman — OpenAI Co-founder
- Lilian Weng
- Andrew Tulloch
- Barrett Zoph
**Key Products:** Tinker API (LLM fine-tuning platform)
### 2.5 Safe Superintelligence Inc. (SSI)
- **Mission:** Build safe superintelligence — nothing else
- **Founded:** June 2024
- **Funding:** $1B from Sequoia, a16z, DST Global, SV Angel
**Founding Team:**
- Ilya Sutskever (Co-founder) — OpenAI Co-founder & former Chief Scientist
- Daniel Gross
- Daniel Levy
### 2.6 World Labs
- **Headquarters:** San Francisco
- **Founded:** 2023
- **Funding:** $230M from a16z, NEA, Radical Ventures, NVentures
- **Valuation:** $1B+ (Unicorn)
**Founding Team:**
- Fei-Fei Li (CEO) — "Godmother of AI", Stanford HAI Co-Director
- Justin Johnson
- Christoph Lassner
- Ben Mildenhall
**Key Products:** Marble (3D world generation platform)
## 3. AI Infrastructure & Data Companies
### 3.1 Scale AI
- **Headquarters:** 303 2nd Street, Floor 5, South Tower, San Francisco
- **New Office:** 650 Townsend St (170,000-180,000 sq ft) — former Airbnb space
- **Founded:** 2016
- **Total Raised:** $15.9B
- **Employees:** 1,400+
**Services:**
- Data labeling & annotation
- Model evaluation
- RLHF (Reinforcement Learning from Human Feedback)
- Defense & government AI projects
**Key Clients:** Meta, Microsoft, Google, OpenAI, US Army, DoD
### 3.2 Surge AI
- **Headquarters:** 2193 Fillmore Street, San Francisco
- **Founded:** 2020
- **Total Raised:** $25M (Series B)
**Services:**
- Data labeling for LLMs
- RLHF training data
- Content moderation
- Search evaluation
### 3.3 Databricks
- **New Headquarters:** One Sansome Street, San Francisco (150,000 sq ft)
- **Previous HQ:** 160 Spear Street
- **South Bay Office:** 200 West Washington, Sunnyvale (305,000 sq ft)
- **Founded:** 2013 (UC Berkeley AMPLab)
- **Investment in SF:** $1B+ over 3 years
**Valuation History:**
| Date | Valuation | Funding Round |
|------|-----------|---------------|
| Dec 2024 | $62B | Series J ($10B) |
| Aug 2025 | $100B+ | Series K |
| Dec 2025 | $134B | Series L ($4B+) |
**Revenue:** $4.8B run-rate (55%+ YoY growth)
**Key Products:**
- Data Intelligence Platform
- Agent Bricks (AI agents)
- Lakebase (serverless Postgres)
- Unity Catalog
### 3.4 Fireworks AI
- **Headquarters:** 2317 Broadway St, Redwood City, CA
- **Founded:** 2022
- **Total Raised:** $307M
- **Valuation:** $4B (Oct 2025)
- **Employees:** 148
**Founding Team:**
- Lin Qiao (CEO) — Former Head of PyTorch at Meta
- Benny Chen — Former Meta Ads Infrastructure Lead
**Services:**
- LLM inference optimization
- Model fine-tuning
- Low-latency AI deployment
## 4. Tech Giants — Gig Economy & Consumer Tech
### 4.1 Uber
- **Global Headquarters:** 1655 & 1725 Third Street, San Francisco (Mission Bay)
- **Real Estate:** JV with Alexandria Real Estate & Golden State Warriors
- **Financing:** $500M refinancing (Feb 2025)
### 4.2 Lyft
- **Headquarters:** 185 Berry Street, San Francisco (Mission Bay, near Oracle Park)
- **Current Space:** 170,000 sq ft (renewed Dec 2024)
- **Previous Space:** 419,000 sq ft (reduced post-pandemic)
- **Founded:** 2007 (as Zimride)
- **Employees:** 4,400+
### 4.3 Instacart
- **Headquarters:** 300 Mission Street, Financial District
- **Current Space:** 60,000 sq ft (2 floors)
- **Lease Term:** 9 years (through 2034)
- **Previous Space:** 107,000 sq ft (expanded 2019, did not fully occupy)
## 5. Y Combinator & Startup Ecosystem
### 5.1 Y Combinator
- **New Headquarters:** Pier 70, Dogpatch, San Francisco (moved 2024)
- **Previous HQ:** Mountain View (17 years)
- **Investment per Startup:** $500,000
**Leadership:**
- Garry Tan (CEO & President since Jan 2023)
- Co-founder of Initialized Capital
- Co-founder of Posterous (YC S08)
- Early employee at Palantir
- Stanford BS in Computer Systems Engineering
### 5.2 YC AI Startup Stats
- **AI Startups in SF Bay Area:** 819+
- **Currently Hiring:** 307+
- **Winter 2025 Batch Growth:** 10% per week aggregate
- **AI Code Generation:** 95% of code is AI-generated for ~25% of startups
### 5.3 Notable YC Alumni (SF-based)
| Company | Category | Notable Stats |
|---------|----------|---------------|
| Scale AI | Data/Infrastructure | $15.9B raised |
| Instacart | Delivery | Public company |
| Coinbase | Crypto | Public company |
| Stripe | Fintech | $95B valuation |
| DoorDash | Delivery | Public company |
## 6. Visionary Leaders & Talent
### 6.1 Dario Amodei (Anthropic CEO)
- **Born:** 1983, San Francisco
- **Education:**
- Lowell High School (San Francisco)
- Caltech (undergraduate, physics)
- Stanford University (BA, physics)
- Princeton University (PhD, biophysics)
- **Career:**
- Baidu (2014-2015)
- Google Brain (2015-2016)
- OpenAI (2016-2021) — VP of Research
- Anthropic (2021-present) — CEO
### 6.2 Sam Altman (OpenAI CEO)
- **Education:** Stanford University (dropped out)
- **Career:**
- Loopt (founder, 2005)
- Y Combinator (President, 2014-2019)
- OpenAI (CEO, 2019-present)
- **Recognition:** TIME CEO of the Year 2023
### 6.3 Ilya Sutskever (SSI Co-founder)
- **Education:**
- University of Toronto (PhD under Geoffrey Hinton)
- **Career:**
- Google Brain
- OpenAI (Co-founder & Chief Scientist, 2015-2024)
- Safe Superintelligence Inc. (Co-founder, 2024-present)
### 6.4 Andrej Karpathy (AI Researcher)
- **Education:**
- Stanford University (PhD, Computer Vision)
- **Career:**
- OpenAI (Co-founder, early team)
- Tesla (Sr. Director of AI, 5 years)
- Independent researcher
- **Notable Quote:** AGI is "at least a decade away"
### 6.5 Fei-Fei Li (World Labs CEO)
- **Titles:**
- "Godmother of AI"
- Stanford HAI Co-Director
- Former Google Cloud AI Lead
- **Contributions:**
- ImageNet (revolutionary dataset)
- Computer vision research
- **Current:** World Labs CEO
### 6.6 Mira Murati (Thinking Machines Lab CEO)
- **Previous:**
- OpenAI CTO (resigned Oct 2024)
- **Current:**
- Thinking Machines Lab CEO & Co-founder
- **Funding:** $2B seed at $12B valuation
### 6.7 Garry Tan (Y Combinator CEO)
- **Born:** 1981 (Winnipeg, Canada)
- **Education:**
- American High School (Fremont, CA)
- Stanford University (BS, Computer Systems Engineering)
- **Career:**
- Palantir (10th employee)
- Posterous (co-founder, sold to Twitter)
- Y Combinator (Partner 2011-2015, CEO 2023-present)
- Initialized Capital (co-founder)
## 7. San Francisco Landmarks & Districts
### 7.1 Iconic Landmarks
#### Golden Gate Bridge
- **Opened:** 1937
- **Span:** 1.7 miles (2.7 km)
- **Height:** 746 ft (227 m)
- **Color:** International Orange
- **Recognition:** Wonder of the Modern World
#### Bay Bridge
- **Connects:** San Francisco ↔ Oakland
- **Opened:** 1936
- **Total Length:** 4.5 miles
#### Pier 39
- **Size:** 45 acres
- **Features:**
- 90+ shops
- 12 full-service restaurants
- California sea lions on K-Dock
- 300-berth marina
- Views of Golden Gate Bridge, Bay Bridge, Alcatraz
#### Salesforce Tower
- **Height:** 1,070 ft (tallest in SF)
- **Completed:** 2018
- **Location:** Financial District
### 7.2 Key Neighborhoods
| Neighborhood | Notable Companies/Features |
|--------------|---------------------------|
| **SoMa (South of Market)** | Anthropic, Salesforce Tower, tech startups |
| **Mission District** | OpenAI, xAI (Pioneer Building) |
| **Mission Bay** | OpenAI campus, Uber HQ, Chase Center |
| **Financial District** | Databricks (One Sansome), Instacart |
| **Dogpatch** | Y Combinator (Pier 70) |
| **Embarcadero** | Ferry Building, waterfront |
| **Nob Hill** | Historic cable cars |
| **Union Square** | Shopping, hotels |
| **Chinatown** | Oldest Chinatown in North America |
| **North Beach** | Italian heritage, Coit Tower |
| **Fisherman's Wharf** | Tourism, Pier 39, Aquarium of the Bay |
### 7.3 Major Venues
| Venue | Purpose | Capacity |
|-------|---------|----------|
| **Moscone Center** | Conventions (Dreamforce, Data+AI Summit) | 700,000 sq ft |
| **Oracle Park** | SF Giants baseball | 41,915 |
| **Chase Center** | Warriors basketball, concerts | 18,064 |
| **Golden Gate Park** | Urban park | 1,017 acres |
### 7.4 Transportation
- **BART:** Bay Area Rapid Transit (connects to East Bay, SFO)
- **Caltrain:** Commuter rail to Silicon Valley
- **MUNI:** SF public transit (buses, light rail, cable cars)
- **Ferries:** To Sausalito, Oakland, Vallejo
- **Cruise (Waymo):** Autonomous vehicles
## 8. Education & Research Institutions
### 8.1 UC Berkeley
**Berkeley Artificial Intelligence Research Lab (BAIR)**
- Computer vision
- Machine learning
- Natural language processing
- Planning & control
- Robotics
**Notable Contributions:**
- Apache Spark (Databricks foundation)
- AMPLab (precursor to Databricks)
- Ray (distributed computing)
### 8.2 UC Berkeley Extension
**AI/ML Course Offerings:**
| Course | Format | Topics |
|--------|--------|--------|
| Artificial Intelligence Foundations | Live Online | Deep learning, CNN, RNN, Keras, PyTorch |
| Introduction to Machine Learning Using Python | Hybrid | ML concepts, algorithms, applications |
| Machine Learning and Deep Learning | Online | Apache Spark, TensorFlow, neural networks |
| Machine Learning with TensorFlow | Online | Data mining, forecasting, signal processing |
**Professional Certificate:** Machine Learning and Artificial Intelligence (6-month program with UC Berkeley Executive Education)
### 8.3 Stanford University
- **Stanford AI Lab (SAIL)**
- **Stanford Institute for Human-Centered AI (HAI)** — Co-directed by Fei-Fei Li
- Major source of AI talent for SF companies
### 8.4 UCSF (University of California, San Francisco)
- Medical AI research
- Health data science
- Biotech AI applications
### 8.5 Academy of Art University (AAU)
- Located in San Francisco
- Design and technology programs
- Digital media and animation
## 9. AI Market Analysis & Metrics
### 9.1 San Francisco AI Office Market
```
Current (2025): ████████████ 5M sq ft
Projected (2030): ████████████████████████████████████████████ 21M sq ft
Growth: +320%
```
### 9.2 Venture Capital Flow
**Bay Area Share of US AI VC Funding:** 78.2%
**Major VC Firms Investing in SF AI:**
- Andreessen Horowitz (a16z)
- Sequoia Capital
- Lightspeed Venture Partners
- Founders Fund
- Index Ventures
- ICONIQ Capital
- Thrive Capital
- Khosla Ventures
### 9.3 AI Workforce Projections
| Year | AI Workers in SF | Growth |
|------|-----------------|--------|
| 2024 | ~20,000 | Baseline |
| 2025 | ~30,000 | +50% |
| 2030 | 50,000+ | +150% |
### 9.4 Key AI Technologies
**Large Language Models (LLMs):**
- GPT-4/GPT-5 (OpenAI)
- Claude 4.5 (Anthropic)
- Llama 3 (Meta)
- Grok (xAI)
**Training Techniques:**
- **Pre-training:** Large-scale unsupervised learning
- **Fine-tuning:** Task-specific adaptation
- **RLHF:** Reinforcement Learning from Human Feedback
- **Constitutional AI:** Anthropic's safety approach
**Model Architectures:**
- Transformers
- Mixture of Experts (MoE)
- Reasoning models (o1-style)
## 10. Mathematical Modeling in AI
### 10.1 Transformer Architecture
The self-attention mechanism is defined as:
$$
\text{Attention}(Q, K, V) = \text{softmax}\left(\frac{QK^T}{\sqrt{d_k}}\right)V
$$
Where:
- $Q$ = Query matrix
- $K$ = Key matrix
- $V$ = Value matrix
- $d_k$ = Dimension of keys
### 10.2 Multi-Head Attention
$$
\text{MultiHead}(Q, K, V) = \text{Concat}(\text{head}_1, \ldots, \text{head}_h)W^O
$$
Where each head is computed as:
$$
\text{head}_i = \text{Attention}(QW_i^Q, KW_i^K, VW_i^V)
$$
### 10.3 Loss Functions
**Cross-Entropy Loss (Language Modeling):**
$$
\mathcal{L} = -\sum_{t=1}^{T} \log P(x_t | x_{
State Frequency Memory networks capture multiple time scales through frequency decomposition for forecasting.
Explain predictions by attributing importance to each input feature.
Shap-E generates textured meshes and implicit functions from text or images.
Shared memory provides common workspace for multiple agents to coordinate.
ShareGPT provides dataset of user conversations with ChatGPT for instruction tuning.
Shift operations move feature map regions spatially enabling mixing without parameters.
ShiftNet replaces spatial convolutions with shift operations dramatically reducing parameters.
Quantum factorization algorithm.
Shortage management allocates limited materials across competing demands prioritizing critical needs.
Change requires many small edits.
ShuffleNet uses channel shuffle operations between grouped convolutions enabling cross-group communication.
Side effects are unintended consequences of optimizing specified objectives.
Signed distance functions implicitly represent surfaces as zero level sets of distance fields.
Tensile stress for nFETs.
Silver recovery from photographic and plating processes reclaims precious metal.
Preserve pairwise similarities.
Simple masked image modeling.
Simple Heterogeneous Graph Network efficiently processes heterogeneous graphs through relation-aware message passing.
Component whose failure stops production.
Single source risk arises from dependence on one supplier for critical materials or components creating vulnerability to supply disruptions.
Multiple GPUs in one machine.
Containers for HPC environments.
Fixed sinusoidal position functions.
Use periodic activations for INR.
Classify skin conditions from photos.
SkipNet learns to skip layers adaptively for different inputs reducing computation.
Service Level Agreements define expected supplier performance including delivery times defect rates and response requirements.
Only attend to recent tokens within fixed window.
Attention mechanism that only attends to nearby tokens within a fixed window to handle long sequences.
Sliding window attention restricts attention to local neighborhoods reducing complexity.
Sliding window forecasting maintains fixed-length training history discarding oldest observations.
Train single network supporting multiple widths.
Represent information in discrete slots.
Small language models provide efficiency with fewer parameters.
Generate molecular SMILES strings.
Average gradients over noisy inputs.
Synthetic Minority Over-sampling Technique generates synthetic examples by interpolating between minority class instances for imbalanced data.
Generate synthetic minority examples.
Use SMT solvers to verify properties.
Attention-based meta-learning.
Ensemble from single training run.