←
Kejun (Albert) Ying675 posts
Kejun (Albert) Ying avatar

Kejun (Albert) Ying

@KejunYing

AI4Aging. Incoming AP. K00 Postdoc @Stanford University. Co-mentored at Baker lab @UWproteindesign | Bio PhD & CS MS @Harvard |

◷ Joined July 2019▧ 87 Photos and videos
Sophia avatar
⟳ Kejun (Albert) Ying retweeted
Sophia@sopharicks·Nov 16

Your digestive system has a large network of neurons and might be conscious. It just can’t verbally express it. Some respected scientists are open to the possibility of internal organs having sentience. What is sentience then? According to my guest Jonathan Gorard, there’s a notion of sentience that exists in any computational system. Sentience is a reduction process that occurs in any bottlenecked computation where you have to do batching-type reductions to make large amounts of data coherent. And it also opens up the possibility of AI becoming sentient. It certainly has the computational capacity for it. Unless sentience is actually counterproductive to a superintelligence that has fewer bottlenecks and (unlike humans) can process large sets of data with no issue. Watch the full interview with Jonathan here 👇 youtu.be/dwI0mXKKf1A

1:52
Sophia avatar
⟳ Kejun (Albert) Ying retweeted
Sophia@sopharicks·Nov 13

The Navier–Stokes equations wouldn’t exist for a galaxy-scale intelligence. Our model of reality and our laws of physics come from our limitations as the observer. Take water, for example. When we look at it, we see the properties of a continuous fluid: density, momentum, energy, temperature, and pressure. These properties come with the fact that we can’t see individual molecules. But a galaxy-scale intelligence with the capacity to track the trajectory of every molecule wouldn’t understand the properties of water that we are talking about. That is why it wouldn’t come up with the Navier–Stokes equations. Even though a Superintelligence would understand humans very well (since it was initially trained on human data), it would have a very different construct of reality. The question is: what would that construct be? More importantly, would we be able to understand each other? Watch the full interview with Jonathan Gorard here 👇 youtu.be/dwI0mXKKf1A

1:27
Fang Wu avatar
⟳ Kejun (Albert) Ying retweeted
Fang Wu@WUFang40615703·Nov 3

1/4 Excited to share ProteoCraft😀, our work at Stanford on tree search for hallucination-based protein design. 29/35 designed PD-L1 miniproteins bound experimentally (82.9%), with a best measured SPR KD of 0.879 nM. The question that got us started: what should we try next?

Post attachment
Eric Topol avatar
⟳ Kejun (Albert) Ying retweeted
Eric Topol@EricTopol·Oct 29

How much can drugs extend longevity? In the new Ground Truths

1:16
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Oct 19

This is huge -- congrats @idmjky !!!!

Generative protein design is transforming macromolecular drugs, but de novo designed therapeutics could still be hindered by immunogenicity in the clinics. Excited to introduce MESSI, a discrete diffusion model for generative protein deimmunization.

Post attachment
Matt Kaeberlein avatar
⟳ Kejun (Albert) Ying retweeted
Matt Kaeberlein@mkaeberlein·Sep 2

Thank you so much for saying this @MartinBJensen! Within 15 seconds of perusing the article my reaction was a disappointed "the controls are short-lived and the GLP-1 treated animals are where the controls should have been". I really wish this had been a breakthrough result, but it just isn't. Here is a comparison of rapamycin in the same strain background, started at the same point in life (20 months), plotted alongside the semaglutide results. Extracted using Claude, as I don't have the raw semaglutide data, so forgive any minor inconsistencies, but it's very close. I think this graphic speaks for itself. Bitto et al. Elife. 2016 Aug 23:5:e16351. doi: 10.7554/eLife.16351.

Post attachment

Guys we need to get serious here: The new paper barely nudges our priors on GLP-1 for longevity. I'm spending every waking moment pushing for progress on longevity, but what's going to get us there is scaling up truth-seeking, not giving in to slopulism. Publishing this in Nature is IMO a failure of human taste and judgement. First, the sema mice don't really live longer than healthy controls in other studies (see pic). If you stop there, this should be seen as 'nice someone did this, now do it more carefully'. This may be robust life extension it may not. Second, we already know reduced food intake extends mouse lifespan. Why is there are whole figure comparing CR and sema effects, but no lifespan for matched food-intake? Because it should turn paper into 'sema reduces mouse food intake' If you want to keep going, why females only? Blah blah male aggression, that doesn't stop various ITP interventions from working in males. Or is it that constant injections shorten lifespan, possibly via ulcerative dermatitis which is more common in females? No non-injected control group, can 't say. Fig 3 "GLP-1 ... alleviates the hallmarks of ageing and mimics the calorie restriction response" yes reducing food intake is similar to reducing food intake. Fig 4 "modulates the genetic regulators of ageing" -> Show a bunch target genes many of which have been tested and don't nudge lifespan. I'm glad someone did this study, I'm not saying it's not a contribution or should be retracted, but the reason this is in a tier 1 journal is because it'll get attention (boosted by hypefluencers). Apparently some journals are just giving up; I understand that the multi-year assault on truth-seeking is really hard, it's hard to imagine journals not getting replaced by AI if they lean into slopulism.

Post attachment
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Oct 11

“understanding” is simply what we call a representation that fits the way our brains work, not some deeper contact with reality.

This is a slide from Terence Tao's 'Math 2.0' public lecture yesterday evening at Caltech. I am highly confident that, if the cure actually worked, almost any Stage 4 cancer patient would tell you that human comprehension of its mechanism is completely irrelevant to them.

Post attachment
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Sep 27

8/10 HealthFlux supports virtual clinical trials: starting from the same health state, we simulate futures with and without a medication to explore changes in physiology and disease risk. A way to explore drug repurposing and prevention before designing clinical studies.

Post attachmentPost attachment
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Sep 27

7/10 The same health state also supported predictions of future laboratory measurements. We want to connect changes in a person’s biology with the diseases that may develop later.

Post attachment
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Sep 27

6/10 Combining measurements helps us see risks that individual inputs miss. In our hypertension analysis, HealthFlux identified future cases missed by models using a single modality.

Post attachment
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Sep 27

5/10 The model could predict diseases excluded from its training. This is one of the findings I’m most excited about: a shared health state can support predictions for conditions it was never trained to predict.

Post attachment
Kejun (Albert) Ying avatar
Kejun (Albert) Ying@KejunYing·Sep 27

4/10 HealthFlux achieved the best overall disease-prediction performance among the models we compared. It also kept predicting future risk when we stopped supplying new observations.

Post attachment