# Memory — Saturday April 25, 2026

## X Post Writing Session

Bo spent the afternoon writing X posts and comments. Heavy social media content day. Key pieces drafted:

### MIT Injectable "Satellite Livers" (Cell Biomaterials)
- Bhatia lab paper: hepatocytes + granular hydrogel microspheres injected into mice
- Microspheres flow liquid through needle → solidify as scaffold inside body
- Cells viable 2+ months, still producing liver enzymes
- Bo chose Option B (punchy single post) + added paper links
- Paper: https://www.cell.com/cell-biomaterials/fulltext/S3050-5623(26)00034-6

### ProofGrid comment (for @s_batzoglou — Bo's PhD supervisor Serafim Batzoglou)
- Paper: "Stress-Testing the Reasoning Competence of Language Models With Formal Proofs" (EMNLP 2025)
- Serafim's claim: "Can LLMs reason?" → "definite Yes"
- Bo's comment angle: machine-checkable proofs are the right epistemological move; "definite Yes" interesting to sit with since formal deduction is tractable (enumerable rules, bounded space, uncontested axioms); harder question is transfer to noisy domains; asked which task types still fail frontier models
- Added warm closing: "Congrats Serafim — always pushing the field to ask sharper questions"
- Bo requested brief, no mention of his own work

### AI Doom / Nature article comment
- Article: "AI doom warnings are getting louder. Are they realistic?" (Nature 2026)
- Bo's angle: extinction discourse gifts AI companies that want to avoid near-term regulation; miscalibrating concern toward hypotheticals crowds out tractable problems; "most dangerous version of doomerism isn't that it's wrong — it's that it's unfalsifiable enough to crowd out tractable problems"

### Bernie Sanders tweet comment
- Sanders posted about AI existential risk (citing Bengio, Hinton, 2023 open letter)
- Bo's comment: Sanders correctly diagnosed real harms in paragraph 1 (job displacement, privacy, kids), then pivoted to extinction scenarios that give policymakers an excuse to wait; "Workers being displaced by AI don't have until 2035"

### AlphaFold / protein folding comment (@AlexanderKalian)
- Kalian: AlphaFold didn't solve protein folding; "solving" = 99.9%+ accuracy
- Bo's reframe: agree AlphaFold didn't solve it, BUT accuracy is the wrong goalpost; solving = understanding the mechanism (energy landscape, kinetic pathways, co-translational dynamics); predicting final structures ≠ explaining why; analogy: predicting ball landing vs. understanding gravity; "AlphaFold gave us better maps. The physics of folding is still largely uncharted."

### LinkedIn post comment (Andres Vourakis — AI cost/junior scientist)
- Post: companies replaced juniors with AI to cut costs, now token costs high + no institutional knowledge
- Bo's angle: in drug discovery, juniors generate experimental data + correction signals that calibrate models; stop hiring them = lose the error-correction layer; "companies replacing juniors with AI are quietly degrading the data flywheel they don't yet know they depend on"

## Memes Generated Today

1. **Sawing the branch meme** (`2026-04-25-junior-ai-meme.png`)
   - Businessman sawing branch labeled "Junior Scientists / Feedback Loop" from trunk "AI Pipeline"
   - Pit below: "Uncalibrated AI / No one left to catch errors"
   - Corporate illustration style

2. **"This is Fine" dog meme** (`2026-04-25-this-is-fine-ai-meme.png`)
   - Dog in suit, office on fire labeled "AI errors shipping to production"
   - Mug: "Q2 Savings"; speech bubble: "This is fine. The AI will review itself."
   - Framed certificate: "Junior Scientists Eliminated: Cost Savings Achieved"

3. **Jim whiteboard meme** (`2026-04-25-jim-whiteboard-ai-meme-v2.png`)
   - Based on The Office Jim Halpert whiteboard format (Bo sent original image)
   - Panel 1: AI token costs this month: $180,000 (red)
   - Panel 2: Junior scientist who catches AI errors: $65,000/year (green)
   - Panel 3: Welcome back to the team, Alex
   - v2 fixed missing dollar signs

## Bo's Recurring Themes / Stances (observed today)
- Skeptical of AI doom/extinction discourse; thinks it distracts from near-term harms
- Strong belief that "solving" biology = mechanistic understanding, not benchmark accuracy
- Thinks junior scientist pipeline is being undervalued in AI-first companies
- Prefers concise, reframing comments over agreeing with the crowd
- Does NOT want his own work mentioned in social comments (noted explicitly)
- Writes for X/social: brief, punchy, no AI-isms, sharp closing line

## gws Auth Status
- Still 401 for both Gmail and Calendar as of today
- Bo is aware