MLE with a focus on Computer Vision. Patent-holder. Inventor of SOTA algorithm. Sick with Long COVID since September 22.
Joined August 2025
- Tweets1.3K
- Following993
- Followers238
- Likes6.1K
Lawrence retweeted
如果 AI 不是泡沫，人们会丢掉工作。如果 AI 是泡沫，人们会丢掉积蓄。很遗憾，最有可能的是，AI 目前处于金融泡沫，但技术上仍处于指数改进曲线上，所以人们会先丢掉积蓄，然后丢掉工作。😇😇😇
Lawrence retweeted
AI bros talking about AI before yesterday’s election vs after yesterday’s election
The differing perceptions of AI are wild. The country just voted against data centers on 40+ provisions nationwide and hates AI. (Plenty of legitimate gripes.) But most of those people are just unaware that AI is finding a new cancer cure or treatment weekly, going through decades of peer reviewed data and uncovering mountains of errors, that its correcting, throughout every academic field. Replacing billions of man hours of drudgery. Closest thing to a “miracle” I’ve ever witnessed. When I talk to people outside of tech/VC, they’re just unaware of all of it.
Lawrence retweeted
As a biologist working in drug development, I’m sick and tired of “curing diseases” being paraded around as the one thing that makes AI worthwhile. AI has not developed a single drug. Nor has it made any unique discoveries that could directly lead to the development of one.
Lawrence retweeted
Als Immunologe kann ich das nur bestätigen: ME/CFS ist eine ernsthafte Erkrankung mit nachweisbarer Immundysregulation, keine bloße Erschöpfung. Dass Betroffene solche Fehler ständig selbst korrigieren müssen, ist wirklich ein Armutszeugnis. #MEcfs
Das ist der Kenntnisstand einiger Journalist*innen, Onlineredakteur*innen, Medienschaffenden etc. zum Thema #MEcfs. Und zwar der Kenntnisstand jener, die darüber schreiben. Im Jahr 2026.
Es kann doch nicht sein, dass es immer wieder die Kranken sind, die diese Fehlinformationen korrigieren. Was für ein Armutszeugnis.
ALT MDR AKTUELL · Folgen 5 Std. Etwa 650.000 Menschen leiden in Deutschland an ME/CFS, auch Fatigue genannt. Typisch dafür ist, dass schon geringe körperliche oder geistige Anstrengungen die Beschwerden verschlimmern können. Hausärzte sind mit den schwer erkrankten Patienten häufig überfordert – auch weil es viel zu wenig Spezialambulanzen gibt. Mehr erfahren über den Link in den Kommentaren. VIELE FATIGUE-PATIENTEN UNZUREICHEND VERSORGT »Was ich mir besonders wünsche, ist, dass Ärzte nicht mehr sagen: 'Wir sind nicht zuständig oder wir kennen uns damit nicht aus.'« Carmen Scheibenbogen Immunologin Bildbeschreibung: Es handelt sich um einen Screenshot eines Facebook-Beitrags von MDR AKTUELL (mit blauem Häkchen und „Folgen“-Button), der vor etwa 5 Stunden gepostet wurde.
Lawrence retweeted
So incredibly proud to finally share Science’s Last Exam (SLE). 🧪✨
We wanted to know: Can AI actually do science? So we gave them real lab tools and hidden expert checks.
The results are a massive reality check. Science is hard, and true end-to-end research is still at 0.
Deeply grateful to the brilliant minds at 34 institutions who made this possible.
🔗sciences-last-exam.org/
The real journey starts now. 🚀
Dive into the thread 👇
I’m incredibly proud to share what we’ve been building at the AItonomy Foundation: Science’s Last Exam (SLE). 🧪✨
Can AI do science? Instead of Q&A, we give agents actual research goals and real-world tools (lab data, software), then grade them against hidden expert checks.
Deeply grateful to the brilliant minds at 34 institutions (Harvard, MIT, Stanford, etc.) who helped craft 113 tasks across 9 disciplines. ❤️
The results? Science is hard. Of 8 frontier models, the best scored 37.5/100. On true end-to-end research? All scored 0. We have an exciting journey ahead! 🚀
(Claude Opus 5.5 & GPT-6 Astra coming soon—just clearing bio safeguards).
Everything opens Nov 6. 🗓️
🌐 sciences-last-exam.org
💻 github.com/Gen-Verse/Science…
🤗 huggingface.co/datasets/Gen-…
Thread below on how the exam works and where models break. 👇 (1/14)
#AGI #AIforScience #AIAgents #SciencesLastExam #OpenAI #Anthropic
Lawrence retweeted
Response to the recent criticism of my work with @resiapretorius on #microclots for the benefit of #pwLC and #pwME dbkgroup.org/a-misdirected-a…
Lawrence retweeted
"We're gunna cure all diseases in 5-10 years"
"Prove it"
"Here's what a bacteria protein looks like. We don't know what it does"
We’re sharing how our biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life.
Vault particles are a longstanding mystery. They are barrel-shaped objects, among the largest protein structures found in our cells. They were first discovered by Leonard Rome and Nancy Kedersha in 1986, but their function remains unknown.
With Claude, we’ve found results that might bring us closer to solving that mystery.
Until now, vaults were only known in the cells of complex things like animals and plants. We have shown that similar proteins from bacteria (and the viruses that infect them) can also assemble into vaults. Finding the structures in these simpler forms of life may make it easier to investigate their function in our own cells.
For our life sciences team, the project was also a demonstration of the potential (and the limitations) of using AI to accelerate science.
In our full piece, we discuss how Claude was able to dramatically speed up some stages—manually determining the vault particle’s structure would have taken weeks to months of painstaking work, but here it took days. We also discuss how Claude made some simple mistakes along the way, and how we collaborated with our machine learning researchers to address them.
Read more: anthropic.com/research/inves…
Lawrence retweeted
Our security researchers built an autonomous cyber vulnerability discovery, exploitation, and patching pipeline.
In a single run with an upcoming cyber model, it validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel at a total API cost of $1.2 million.
Lawrence retweeted
regular reminder that almost any cell (besides certain fully diffed b cells and neurons) can be an antigen presenting cell if you hit it with enough ifny
Astrocytes are fully competent APC but only conditionally: IFN-γ & microglial cytokines are jointly required for astrocytes to drive Ag-specific CD4+ T cells @CellReports
cell.com/cell-reports/fullte… preview @FChrisBennett
cell.com/cell-reports/fullte… @LiddelowSA @borderbiologist @NYU_CNS
Lawrence retweeted
Through work, one develops as a person, provides for family needs, expresses one’s own talent, participates in social life, and builds a better world by collaborating in God’s creative work. The real challenge is to offer everyone opportunities to make the most of their abilities. Unemployment, exploitation, and working conditions that endanger health and life itself are unacceptable (Gaudium et Spes, 67). #GeneralAudience
Lawrence retweeted
It's important not to get so focused on what's happening in someone's brain that we forget to ask what's happening in their life.
Lawrence retweeted
Many people hope AI will one day cure diseases we’re still powerless against. So do I. To me, that would be AI’s greatest contribution to humanity.
The leading AI labs have bold visions:
🔬 Dario Amodei believes future powerful AI could compress 50–100 years of progress in biology and medicine into 5–10 years.
👉 “Machines of Loving Grace”
🧬 Demis Hassabis hopes AI could eventually help cure all diseases.
👉 His CBS “60 Minutes” interview
I wanted to track what becomes of these promises. So I had my AI agent build "Are We Cured Yet?": a website that compares verifiable medical progress with the big announcements.
The counter for proven cures and therapies developed using AI currently still stands at 0.
There’s a long road between a research breakthrough and a cure. I really hope we’ll see that counter rise soon.
👉 AreWeCured.com
✖️@AreWeCuredYet
What do you think is hype, and what’s realistic? And which medical breakthrough from AI are you personally hoping for?
Lawrence retweeted
J’ai déposé une proposition de loi pour en finir avec l’invisibilisation et l’abandon des malades du Covid long !
Les chiffres sont inquiétants : en 2022, Santé publique France recensait 2 millions de malades, dont 700 000 atteints de formes sévères. Une personne atteinte sur cinq cesse de travailler, et un grand nombre de celles qui poursuivent leur activité ne peuvent reprendre qu’à temps partiel ou finissent par être licenciées. Des jeunes sont radiés de leurs établissements d’enseignement, et les enfants malades restent sans repérage à l’école, faute de formation des médecins scolaires.
Voilà ce que contient ma proposition de loi 👉
Lawrence retweeted
T細胞エンゲージャーの標的が、がん細胞から離れた。
GenentechのグループがScience Translational Medicineで報告したのは、CCR8とCD3を同時につかむ二重特異性抗体。狙うのは腫瘍そのものではなく、腫瘍の中で免疫を抑えている制御性T細胞。
論文自身が書いているとおり、T細胞エンゲージャーはこれまで腫瘍特異抗原を介してがん細胞を狙う用途に集中し、免疫抑制細胞の除去にはほとんど使われてこなかった。
理由づけが明快。抗体単剤でCCR8陽性の制御性T細胞を消す方法はNK細胞に依存する一方、固形がんではNK細胞が乏しいことがある。T細胞エンゲージャーなら抗原特異性を上書きして、その場にいるT細胞を動員できる。
結果はヒト固形がん検体での除去、複数のマウスモデルでの腫瘍増殖抑制、抗PD-L1との併用での増強、カニクイザルでの良好な忍容性。いずれも前臨床で、ヒトに投与した試験はまだない。
腫瘍を殺す道具から、抑える側を外す道具へ。同じ仕掛けの向きが変わった。
※2026年10月28日掲載。著者は全員Genentechの所属。前臨床のみで、用量や腫瘍縮小率などの数値はAbstractに記載がない。私の補足として、制御性T細胞を減らす設計は原理的に自己免疫の毒性と背中合わせで、サルでの長期所見はAbstractに示されていない。
出所：Science Translational Medicine 18(873)（2026/10/28）T cell–mediated depletion of intratumoral regulatory T cells using a CCR8xCD3 T cell–engaging bispecific antibody
doi.org/10.1126/scitranslmed…
@ScienceMagazine
Lawrence retweeted
“Before birth, injuries heal without inflammation and don't leave a scar. Two important causes for that difference are the high concentration of CO2, which limits lactic acid production, and the absence of the n-3 and n-6 polyunsaturated fatty acids. In adulthood, the tissues become progressively more loaded with fats of that type, leading to greater production of inflammatory agents such as prostaglandins, and an increasing tendency to produce lactic acid rather than CO2.” — Ray Peat
“The results of our research confirmed that transcutaneous application of gaseous CO2 significantly improved the healing of chronic wounds in diabetic patients. Wounds in study group patients that received CO2 therapy healed significantly faster compared with the control group. After CO2 therapy, 66% of the wounds healed completely compared with 0% in the placebo group.” teapawt.net/lowmegatron/status/187…
Ray Peat and Bud Weiss - The Biology of Carbon Dioxide
Better audio, brightened the video.
00:00 Language and altitude, a trip to Russia, Ray’s introduction to science
01:00 J.C. Bose and the properties of life
04:00 Twitching nerves and rocks
05:50 Respiration, Otto Warburg, Albert Szent-Györgyi, William Blake, Swedenborg, brain physiology
07:00 Getting a degree in biology by keeping quiet, membranes
08:30 Gilbert Ling, surface electrical effect
10:00 Science, money, and prestige
11:00 W.F. Koch, respiration, cancer treatments, the quinone system
13:00 Contraction, respiration as electrical property, cardinal adsorbents
14:00 What is CO2?
15:30 Lewis acids, electron donors and acceptors
1700 CO2 as a Koch reagent
18:30 Stabilizing the system, cardinal adsorbents - CO2 and progesterone
19:30 Protein conformation, sodium and potassium
21:00 Hair ion exchange, membrane pumps
22:30 Buteyko, oxygen, and the Bohr curve
23:30 CO2, calcium, and bone density
26:00 Energy at altitude
27:30 Osteopetrosis, marble bone disease
28:30 CO2 baths for cardiovascular disease (watching TV)
29:30 The essentiality of CO2 for all life
30:30 Ideal environmental CO2 levels, planetary temperature
31:00 Low light environments, CO2 and life
32:00 Loss of CO2 with aging, frogs and salamanders
33:30 CO2 protects from poisoning and hypoxia
34:30 The naked mole rat
35:30 Queen bees, lipid peroxidation, polyunsaturated fatty acids, and longevity
36:00 CO2 lowers lipid peroxidation
36:30 Near-death experience, NDE and CO2
37:00 Pure oxygen, medical death, permissive hypercapnia, shrinking the brain
38:30 Stroke, transient ischemic attacks, Coke, baking soda, paralysis
39:50 Curing septic shock and loss of circulation with CO2
41:00 Carbogen, post-war, reductionist medicine
42:15 Hyperbaric oxygen therapy, CO2, water, and cancer
44:00 Absorbing CO2, bones and protein synthesis, bag breathing, and blood pressure
45:00 Bats, caves, longevity and high metabolism
46:00 Dropping a tank of CO2, arthritis
47:30 Breathlessness, Co2 stimulating its production, ETC, cytochrome oxidase, thyroid, altitude, mitochondria
49:30 Diabetes, cancer, lactic acid, NAD, NADH excess, inflammation, cell pH
51:00 CO2, lipolysis, glycolysis, free fatty acids, respiration
52:30 Acid/alkaline, water economy, electrons, gelatin, mechanical hyperventilation
55:00 Altitude, pollution, and asthma
56:00 Did you say you sit in a plastic bag full of CO2?” CO2 springs, membrane gradients
58:30 The Bohr effect
59:00 Carbaminos, pituitary hormones, prolactin, growth hormone, and CO2
1:01:00 So-called receptors
1:02:00 Leaf bag full of CO2
1:03:00 Henderson–Hasselbalch equation, acidifying the system, kidneys
1:04:30 What is the role of bicarbonate in acid/base regulation?
1:05:30 Pregnenolone, progesterone, estrogen, and lactic acid
1:07:00 Thyroid hormone (T3) CO2, respiration, calcium carbonate in bones
1:08:30 Endotoxin, permeability, nitric oxide, TNF, estrogen, suppressing respiration, hypothyroidism
1:10:00 Isn’t estrogen good for the brain? Coke (the other one) is safer
1:12:00 Mae Wan Ho, polarisation streams
1:13:00 The living state, reading a newspaper through a fish
1:14:00 Life as a liquid crystal
1:15:30 CO2 as a context for system models, the limits of reductionist science
mp3 link below
Lawrence retweeted
If a healthy person felt how I feel for one hour they would think "you couldn't possibly live like that"
Lawrence retweeted
CD19-CAR T cells clear the EBV-infected B cells that carry autoreactive antibodies. Rituximab doesn’t. That may be why only one of them produces lasting remission.
New Robinson Lab preprint led by me and @panditmahes. 🧵
biorxiv.org/content/10.64898…
Lawrence retweeted
>be cell biology postdoc at UCLA, 1986
>job: purify vesicles from rat liver
>ez
>purify the vesicles
>something else in the tube
>weird little barrels, way bigger than a ribosome
>fkisdis
>purify the barrels instead
>mostly protein, tiny bit of RNA
>not related to vesicles at all
>new organelle just dropped?
>lab holds a naming contest
>"vaults" "sure, "great name" "
>write it up, publish
>conclusion section: "function: unknown"
>surely we'll figure it out in a couple years
>reviewers: "are you sure this isn't just a virus"
>prove it isn't a virus
>apply for grants to study it
>first question on every call: "what does it do"
>"we're working on it"
>be early 90s
>vaults hang around the nucleus
>"they must plug the nuclear pores"
>they do not
>be different scientist, Netherlands, 1995
>chasing LRP, mystery protein that shows up in chemo-resistant tumors
>clone it
>~88% identical to rat vault protein
>it was vaults the whole time
>whole cancer field piles on
>a decade of papers on vaults and drug resistance
>nobody ever proves vaults cause it
>back at UCLA
>one of the vault proteins binds telomerase
>"guys what if this is the fountain of youth"
>knock it out in mice
>telomeres fine, aging fine, mice fine
>2002
>knock out the main vault protein
>mice: perfectly healthy
>knock out all three vault proteins
>completely vaultless mice
>mice: basically fine, some subtle immune stuff
>mfw something in every cell since the dawn of eukaryotes does nothing obvious
>"just test it in yeast"
>yeast don't have vaults
>"worms then"
>no vaults
>"fruit flies"
>no vaults
>mfw every easy model organism dropped them
>try making recombinant vaults in insect cells
>expecting to get some loose protein at best
>get fully assembled vaults
>can't find a single free copy of the protein anywhere
>turns out ribosomes 3D-print the barrel straight off the mRNA
>lolwut
>still no function
>be structural biologists
>spend years modeling it as 96 proteins with 8-fold symmetry
>2009 X-ray structure
>it's 78 proteins, 39-fold
>caps still blurry
>2025: caps are closed, 13-fold, tiny hole in the middle
>still no function
>NIH funds vault research for 15 years
>NIH: "so what does it do"
>"...we're working on it"
>NIH stops returning calls
>if we can't figure out what it does, we'll make it do something
>stuff a cancer-fighting chemokine inside the barrel
>inject into mouse lung tumors
>immune cells swarm in, tumors shrink
>2011: "human trials within three years"
>2024: "human trials next year"
>2025 database: preclinical
>start a YouTube channel explaining vaults to anyone who'll listen
>be different scientist, Broad Institute
>want to record what genes a cell turned on over time
>try tethering mRNA to a bacterial protein
>doesn't work
>stumble onto a YouTube channel about some obscure barrel organelle
>engineer vaults to grab mRNA
>RNA half-life goes from ~17 hours to ~132 hours
>publish in Science, 2026
>use it to catch drug-resistance genes in lung cancer cells before the drug even hits
>the vault guy is thrilled
>still nobody knows what vaults are for
>be epigeneticists
>screen the whole genome for methylation set at random in the early embryo
>separately study kids conceived in different seasons in rural Gambia
>both screens independently point at the same gene
>it's the goddam vault RNA gene
>its methylation depends on conditions around conception, locked in for life
>linked to cancer outcomes, blood sugar, cleft palate
>mfw the mystery barrel has a mystery RNA that remembers when you were conceived
>40 years later
>one of the most abundant structures in your cells
>still not in a single textbook cell diagram
>we use it anyway because yolo
>function: unknown
>be CNRS scientist in montpellier
>dig through institut curie's chemical library looking for an HIV cure
>find a quinoline that blocks rev from exporting viral RNA
>somehow involved in splicing
>start a company called splicos based on this
>4 years later, company gets merged into abivax
>molecule is now ABX464
>run two phase 2a HIV trials
>it shrinks the viral reservoir a bit
>mfw "a bit" does not cure HIV
>but the bloodwork is weird
>random microRNA miR-124 is cranked tf up
>mfw miR-124 is a major chill pill for inflammation
>turns out the drug grabs the cap-binding complex and splices exactly one random lncRNA
>that lncRNA becomes miR-124
>why that one? lol idk
>HIV attacks GI tract, so maybe give it to colitic mice?
>guts heal lmao
>let's yolo this for ulcerative colitis
>2018 phase 2a works
>let's try with rheumatoid arthritis while we're at this
>2020 covid hits
>"what if it stops cytokine storms"
>yeah let's cure ARDS too lol
>french gov makes it a national research priority
>500 patients in
>2021 trial stopped, does nothing
>stock eats dirt
>ceo: "covid was hyper-acute, chronic inflammation is different, trust me"
>quietly drop rheumatoid arthritis
>go all in on UC
>phase 2b works
>2025 phase 3 induction hits
>stock goes vertical
>2026 maintenance trial: half get remission on drug vs 10 % on placebo
>placebo rate is the lowest ever seen in a trial like this
>NDA to FDA planned late 2026
>mfw a failed HIV drug, a failed covid drug, and a lucky lncRNA might become a blockbuster UC pill
>mfw it took 18 years
Lawrence retweeted
It’s actually worse to have extreme AI hype in a few ways:
1.) Irresponsible to patients to give them false hope.
2.). Billions of dollars gets allocated to dumb things*
3.). Translational drug candidates that work in animal studies don’t even get considered for investor funding b/c they don’t have an “enough AI.”
Is the goal to make drugs or “cures” or to use AI? Currently a lot of the hype is just to use AI with the assumption, but not much evidence it actually leads to more drugs that work and are safe in humans.
I’m pro-AI and use AI all the time. I have accelerated my R&D immensely.
* which will not meaningfully increase # of drugs or “cures” in the near-term or maybe ever.
