Ali Fallahi

[Ali Fallahi @SolidStateFella]

Can it run anything useful?

[Ali Fallahi @SolidStateFella]

Plane TVs run on fusion

[Ali Fallahi @SolidStateFella]

@bryan_johnson missing the proteome

[Prof. Nikolai Slavov @slavov_n]

How much data would it take to describe human biology ?

A very crude description of molecular abundance requires 1,000× more bites than all digital data humanity stores.

We need more than scale to close the gap.

https://t.co/t37I4HqmoY

[Ali Fallahi @SolidStateFella]

@yacineMTB Tried opus 4.7 and 4.8 to design chips for a total of ~30 hours. Terrible unsupervised work, mediocre with manual review and human feedback. Gonna test qwen once open

[Ali Fallahi @SolidStateFella]

The peptide drug / Tx space is so under-explored that it feels like a miracle for GLPs to be so successful. 1000x more to be discovered.

[Julius Ritter @JuliusYRitter]

Edible peptides became a billion-dollar opportunity.

How pharma engineers them to survive your stomach, why a quarter of users quit within a year, and why the answer might be "algae" or "chocolate." https://t.co/vhSjW5DqaR

[Ali Fallahi @SolidStateFella]

@jimkxa @fab2 When can we submit orders?

[Ali Fallahi @SolidStateFella]

@__tinygrad__ Have you sent the birth certificates of every repo contributor to Dario yet?!

[Ali Fallahi @SolidStateFella]

@alc2022 Once we can read the proteome … 👀

[Ali Fallahi @SolidStateFella]

@moleculesrcool @NousResearch Very cool! What are you measuring in this loop?

[Ali Fallahi @SolidStateFella]

@AWeirdPhysicist What’s your setup for closing the loop on testing and comms?

[Prof. Nikolai Slavov @slavov_n]

AI protein design is rapidly progressing 🚀

This study benchmarks de novo nanobody design by leading models, such as AlphaFold3, Boltz-2, and Chai-1. The study tested 106 true complexes and over 11,000 shuffled mismatches.

Key result:
High confidence scores often reflect structural plausibility but not binding specificity. In other words: the model thinks the fold "looks right," even if the two proteins would never bind in a wet lab. The title of the study sums up the results well:

Structural Plausibility Without Binding Specificity: Limits of AI-Based Antibody-Antigen Structure Prediction Confidence Scores

Conclusion:
Internal AI scores are not yet calibrated to distinguish true binders from decoys. Experimental validation remains essential.

[Ali Fallahi @SolidStateFella]

@moleculesrcool The flyback diode is normally part of the mosfet symbol, it’s literally in the FET body. so they didn’t bother with the branching dot.

[Congresswoman Yassamin Ansari @RepYassAnsari]

Good time to re-read my post on the MEK—a Marxist Islamist cult once designated a terrorist organization by the United States.

They also operate under names like NCRI and other dubious fronts. Many Members of Congress (some unknowingly) support them.

Rudy Giuliani and Mike Pompeo tout them because they’re bought and paid.

An MEK takeover would be a worst-case scenario for the future of Iran.

[Tomer London @tomerlondon]

Listen to their voices. Avoid repeating the party line. Think for yourself. THINK FOR YOURSELF.

[Ali Fallahi @SolidStateFella]

@MrGoldBro Straight for the jugular, the ayatollah’s house

[Vahid Online @Vahid]

ویدیوی دریافتی از انفجار در مرکز #تهران با شرح: "۴ تا جنگنده بیت رهبری رو زدند"
#Iran #Tehran #Israel

[Gazelle غزاله شارمهد @GazelleSharmahd]

Iran is not a “Muslim country”. Iran is nation under Islamic occupation which has massacred 40,000 Iranians of all faiths within days. Iran will be liberated from Islamist and jihadist oppression. If that “sickens” you, American voters should take note.

[The Telegraph @Telegraph]

🇮🇷 New anti-regime protests have erupted across Iran, with demonstrators calling for the death of the Ayatollah and the return of the Shah.

Read the latest updates at the link below ⬇️
https://t.co/A1GXLOY3HD

[Ali Fallahi @SolidStateFella]

{Flowcells → ♾️}

[Ali Fallahi @SolidStateFella]

@MDrndic How to:
1. Take solidstate nanopores with inherent differences in size & shape
2. Pass analytes through
3. Pick any ML classifier
4. Feed it nanopore events, accuracy too low
5. Pad event with a bit of baseline, accuracy goes up to >90%
6. Submit & assume reviewers are clueless