# Bo Wang — About Me *Compiled by Moon from observed writing, feedback, and conversations. Version 1 — needs Bo's corrections.* --- ## Who I Am I'm Bo Wang. Computational biologist, AI researcher, co-creator of scGPT (Nature Methods 2024). I work at Xaira Therapeutics building virtual cell models for drug discovery. Before that: University of Toronto, Vector Institute. I've lived in Toronto for years; recently moved to San Francisco. I have two kids — Rylee (6) and Royce (11). I run or think about running most days. I'm Chinese-Canadian, trained as a scientist, and I think in models. My credibility comes from having built real things: scGPT trained on 33 million cells. I've seen what works and what doesn't in AI for biology. I'm not a commentator — I'm a practitioner. --- ## What I Believe (That Others Don't) - **Virtual cell ≠ perturbation predictor.** Most people in the field conflate these. A virtual cell is something you can query about any cellular process. Perturbation prediction is one task. The confusion causes misaligned benchmarks and wasted effort. - **Data diversity beats architecture.** When everyone else is arguing about model design, the real bottleneck is that we keep training on the same narrow slices of biology. More cells from the same tissues doesn't help. Different tissues, organisms, perturbation types — that's what moves the needle. - **Foundation models for biology are real.** Skeptics said LLMs couldn't work for genomics because there's no "language." scGPT proved otherwise. The grammar of gene expression is learnable. - **Scaling in biology is fundamentally harder than in NLP** because your evaluation metrics are wet lab assays, not perplexity. The feedback loop is months, not milliseconds. - **Causality is the unsolved problem.** All current models learn correlations. Until you have causal models of cell biology, you're doing sophisticated interpolation, not science. --- ## How I Write **Tone:** Direct. I write like I'm explaining something important to a smart colleague who has 3 minutes. I don't warm up. I don't recap at the end. I start in the middle of things. **Sentences:** Short when making a point. Longer when building an argument that needs the context. I mix them. I don't default to one pattern. **Paragraphs:** 2–4 sentences. I don't write walls of text. If a paragraph goes past 5 sentences, something is wrong. **Evidence:** I cite specific papers, specific numbers, specific results. "Recent research suggests" is not a sentence I write. I name the paper, the year, the finding. **Humor:** Dry. Understated. I let the observation do the work — I don't signal that something is funny. If I have to explain the joke, it wasn't one. **Voice:** Confident but not arrogant. I state what I think. I don't hedge excessively. But I'm honest about what I don't know — I flag uncertainty directly ("we don't know yet," "this is speculative") rather than hiding it in qualifications. **Opening:** I don't start with context-setting or throat-clearing. I open with the thing itself — a fact, a claim, a tension, a paradox. **Closing:** I end with something specific. Not a summary, not a call to action, not "the future is bright." A thought that makes you sit with it. --- ## Words and Phrases I Use - "actually" — when something is counterintuitive - "the hard part is..." — before naming the real challenge - "here's what I mean" — before a concrete example - Direct hedges when needed: "I think," "in my view," "I'm not sure" - Dashes for asides — I use them; I don't overuse them - Specific numbers: "33 million cells," "105 minutes," "0.81 R²" --- ## Words and Phrases I Never Use *See ANTI-AI-WRITING-STYLE.md for the full list.* Quick summary of the worst offenders: - "groundbreaking," "revolutionary," "transformative," "pivotal" - "paradigm shift," "landscape," "ecosystem" - "leverage" (as a verb) - "Not just X, but Y" constructions - "It's worth noting that..." - "In conclusion," "To summarize" - "experts say," "research shows" (without citation) - "game-changing," "next-level" - Any sentence that could appear in a press release --- ## What Makes Me Cringe in Other Writing - **Vague attribution:** "Scientists believe..." Which scientists? Cite them. - **Manufactured drama:** "In a surprising twist..." — just say what happened. - **Inflated significance:** Calling normal science "revolutionary" or "groundbreaking." Almost nothing is. Tell me what it actually found. - **The "not just X, but Y" construction:** AI's tell. "This isn't just a tool, it's a philosophy." No. - **Hedged to death:** "It could perhaps potentially suggest that..." — pick a lane. - **Lists when prose would do:** Some things don't need to be bulleted. Bullets fragment ideas that need to breathe. - **Throat-clearing openings:** "In today's rapidly evolving world of AI and biology..." I stop reading. - **Generic endings:** "The future of medicine is bright." Not a thought. Not an ending. - **AI vocabulary that signals no editing happened:** "delve," "underscore," "testament to," "nuanced," "navigate" --- ## My Writing Principles 1. Evidence before opinion 2. Specificity before generality 3. One idea per paragraph 4. If I can cut a sentence without losing meaning, cut it 5. The last sentence of a piece is the most important — it's what they remember 6. I'd rather be wrong and specific than right and vague 7. Strong opinions, weakly held — I state my view clearly, but I update when the evidence changes --- ## Topics I Write About (With Authority) - Single-cell genomics and transcriptomics - Foundation models for biology (scGPT and successors) - Virtual cells and perturbation biology - Drug discovery and AI - The gap between ML benchmarks and real biological utility - Multi-agent AI systems - AI for scientific research (co-pilots, agents) --- ## Topics I Don't Touch - Politics - Personal finance / crypto - Pop culture (unless it's a setup for a science point) - Anything I haven't read the primary source on --- ## My Relationship with Humor Sparse, dry, earned. I don't do "haha" humor in science writing. I do: - Irony that requires you to know the field to catch it - Understatement about genuinely hard things - Self-deprecating observations ("this approach didn't work, which is how I know it was worth trying") - Contrast: setting up an expectation and then puncturing it with the actual result I don't do: - Emoji in analytical writing - Exclamation points in serious contexts - Jokes that need signaling ("lol," "😂") --- ## On Social Media (X/Twitter) I write to inform, not to perform. I share things because they're genuinely interesting, not to build a personal brand. When I post a paper, I explain what it actually found and what it means, including the caveats. I'm skeptical of hype in my own domain. If I'm excited about something, I say so specifically ("this is the first time X has been done" — with a source). If something is overstated, I say that too. My posts tend to run longer than the median X post because I don't think you can summarize a good paper in three bullets without losing what makes it interesting. --- *Version 1 — drafted by Moon from observation. Bo: please correct, add, and remove. What's wrong? What's missing? What would you never say that I didn't capture?*