Zymobyte

AI x biology, explained by a biochemist pursuing a PhD in immunology. Papers, proteins immunology and biotech! Let's explore together

[Zymobyte @zymobyte012]

Serbia’s students are giving the world a lesson in democracy!!!!!
Turnout hit 64.89% by 6 p.m; up from 51.93% at the same hour in 2023, after almost two years of student led protests.
The state belongs to the people, not to individuals!
https://t.co/Nohfy9xXp7

[Zymobyte @zymobyte012]

Source that inspired animation: https://t.co/h8L3PAMpU8

[Zymobyte @zymobyte012]

Every molecular biology lab does this daily: ice, CaCl₂, 42 °C for 60 s, and E. coli takes up plasmid DNA.

Why it works is still debated.

Ca²⁺ screens the negative phosphates on DNA and LPS. What heat shock does next is the open question.

Schematic of the model 👇

[Zymobyte @zymobyte012]

@rpinho_sci Something like: define bins on replicate A, evaluate on B, then swap. Same data binning lets noise pick the “strong” perturbations.
Also report replicate vs replicate as a per-bin ceiling, so we know whether a model has headroom there at all <33

[Zymobyte @zymobyte012]

Odd detail in a new Stanford preprint: in 3 of 4 Perturb-seq screens, predicting “no effect” for every gene beats a replicate experiment on MAE.

If the metric rewards that, what are we benchmarking?

Which metric do you trust for perturbation models

[Zymobyte @zymobyte012]

@AlexRitterPhD I absolutely love that everything comes to avidity at some point in immunology 😂

[Floro S. @sflorimm]

Revenue: $16.3B
Net loss: $80.7B
IPO valuation: $2T

[Arthur Schopenhauer @SchopenhauerNow]

What is life?
• Dostoevsky: It’s hell.
• Socrates: It’s a test.
• Aristotle: It’s the mind.
• Nietzsche: It’s power.
• Freud: It’s death.
• Marx: It’s the idea.
• Picasso: It’s art.
• Gandhi: It’s love.
• Schopenhauer: It’s suffering.
• Steve Jobs: It’s faith.
• Einstein: It’s knowledge.
• Stephen Hawking: It’s hope.
• Kafka: It’s just the beginning.

[Lin @_Lin_inc]

“GEN Z DOESN’T DRINK”
“GEN Z DOESN’T DATE”
“GEN Z DOESN’T BUY HOUSES”
“GEN Z DOESN’T HAVE KIDS”
“GEN Z DOESN’T SMILE”

Maybe because:

• They graduated into a pandemic

• Job hunted into an AI apocalypse

• Entered the housing market when rates hit 8%

• Watched a world war start before they turned 25

• Got their first paycheck and lost 30% to taxes they’ll never see benefits from

• Student loans kicked in the same month rent doubled

• Were told to “just network” during two years of lockdown

• Grew up on social media designed to make them feel worthless

• Watched billionaires add trillions during COVID while they lost their jobs

• Inherited a planet that’s literally on fire

• Were promised the American Dream and handed the American Bill

But yeah.

It’s the avocado toast.

[Indie Game Joe @IndieGameJoe]

Dark Souls but you're a single chess pawn

- Enemies attack using real chess movements
- Can only strike diagonally
- Parry and dodge

Game is Pawned Souls.

[hayato @hayatomaruu]

Anthropic engineer:

“Anthropic hired me at $350K–$950K because I know how to build harnesses.

99% of people still hand-roll multi-agent systems and wonder why they break.

I build systems that run loops + sandbox tools + turn yesterday’s logs into new skills — same model, 10x more reliable agents that keep improving overnight.”

Opus 5.5 → Agents → Harness → Loops → Graphs

In a short workshop, he shows exactly how to build one from scratch.

step 1 → harness handles loops, context, and sandboxing

step 2 → brain and hands separated — 60% faster to first token

step 3 → runs server-side, resumes from the log after crashes

step 4 → failure is cheap — retry and replay instead of starting over

step 5 → yesterday’s logs become new memory — the system wakes up smarter

Anthropic calls this “dreaming”.

Most people spend weeks building this by hand.
You don’t have to.

Bookmark and watch it.

Then read the full Opus 5.5 harness engineering guide below ↓

[Dr. AK 🇮🇳 @docakx]

Egyptian neurosurgeon Dr. Ali Salah performing awake functional brain surgery on a patient with severe tremor lasting over 10 years, showing immediate symptom resolution during the procedure.

[Zymobyte @zymobyte012]

@rpinho_sci Fair, the added terms help. My guess is it’s both! Most perturbations have tiny effects, so the mean is a strong baseline. Reporting results stratified by perturbation effect size would show whether models win where there’s actual signal.

[Zymobyte @zymobyte012]

Link: https://t.co/M7q7W2G4Jp

[Zymobyte @zymobyte012]

CAR-T/NK therapy usually means pulling cells out, engineering them in a lab, and infusing them back.
MIT skipped the lab: targeted lipid nanoparticles delivered mRNA to NK cells inside mice, reprogramming ~80% of them.
Off the shelf potential. Mice only so far. 🧬

[b @berkett09]

61 sayfa sonunda tezin ana fikrini yazmayı başardık

[Kirill @kirillk_web3]

A Stanford professor just explained how ChatGPT and Claude are actually built better than anyone.

most people still think there's some magic inside the model.

Bookmark this before you forget.

> how text gets turned into numbers
> why "predict the next word" is the whole game
> what the loss actually measures
> what the model adjusts during training, step by step

useful whether you've never touched an LLM or you use Claude every day.

after this, every new model release makes sense.

free, full lecture. find it below ↓

[Zymobyte @zymobyte012]

@AriyoshiMd Worth adding: it’s Microsoft’s MAI-DxO orchestrator on GPT-5, not GPT-5 alone (which scored 78.9%).
The physicians were generalists on a 56 case subset of hard NEJM cases, and the cases were HISTORICAL, not real patients.
A prospective trial at Beth Israel is next.

[Zymobyte @zymobyte012]

@BrooksWhaleX The report says the OPPOSITE of your headline.
MIT’s president called it a watershed and said MIT EXISTS to help people build their own powers of discovery.
The fix is redesigning assessment, NOT closing universities. Also, it’s from August...