Ben Neely

Ben Neely

@benneely.com

Analytical biochemist attempting molecular cartography across the tree of life, powered by MS-based proteomics. Opinions expressed are solely my own. he/him benjaminneely.com Charleston, SC, USA

2K followers771 following3.1K posts

Ben Neely
@benneely.com

Not cool real world effect of AI: new required training likely AI-assisted creation and uses AI voice (reading a script). Way easier to make, right, no limits on voice actor cost, all good right? Nope, this dang thing is freaking 4 hours long. Limits, both monetary and effort were good.

Ben Neely
@benneely.com

What's funny is we had genAI training that was pretty fun. It was like they did The Office, but the actors dealt with genAI. Real humans and fun little acting. This thing is static human pics with voice over. And I am thinking not real humans.

Ben Neely reposted

ypic.bsky.social
@ypic.bsky.social

🎙️ The YPIC Podcast is back! After a short break, we’re excited to announce the return of the YPIC Podcast with a new episode: 🔬 DDA vs DIA: Choosing the Right Strategy for Proteomics For our comeback episode, Lucía Casado and Maico Lechner sit down with three leading experts in proteomics:

Ben Neely
@benneely.com

If you want an easy way to navigate or predict flooded roads, consult old maps. Most “new” places flood, cause why would anyone have built there.

Ben Neely
@benneely.com

If we use peers as “is this legit”, I see why a well-controlled LLM could be better than slow and petty flesh bags… but then, is it just AI submitting and reviewing? 🪦🧪 H/t @michaelhendricks.bsky.social and @richardsever.bsky.social who probably have better threads to see some conversation.

Jan Kabatek 🇺🇦@jankabatek.com

Gary King just dropped a study showing that AI peer reviews now dominate human peer reviews on every metric they tested. gking.harvard.edu/peerless/

gking.harvard.eduAI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices | Gary KingWe show that AI-generated article reviews are now better, by every measure we could construct, than reviews written by well-compensated human experts and judged to be of unusually high quality by jour...
View full thread
Ben Neely
@benneely.com

Wait, is the pushback at its core because we are destroying the currency of science? If it’s not about pubs or solutions, but has left thought and learning (and using your own brain to evaluate each other) minus the incentives, the system won’t know how to function.

Ben Neely
@benneely.com

Does it make a future where after all the cognitive offloading and deskilling that education and research become artisanal pursuits?

Ben Neely
@benneely.com

If we use peers as “is this legit”, I see why a well-controlled LLM could be better than slow and petty flesh bags… but then, is it just AI submitting and reviewing? 🪦🧪 H/t @michaelhendricks.bsky.social and @richardsever.bsky.social who probably have better threads to see some conversation.

Jan Kabatek 🇺🇦@jankabatek.com

Gary King just dropped a study showing that AI peer reviews now dominate human peer reviews on every metric they tested. gking.harvard.edu/peerless/

gking.harvard.eduAI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices | Gary KingWe show that AI-generated article reviews are now better, by every measure we could construct, than reviews written by well-compensated human experts and judged to be of unusually high quality by jour...
Ben Neely
@benneely.com

There are tasks that can be done better but the true loss is training/community. A benign example is game decomps: they want to form groups and pound on a problem and learn how games work (same in math, etc.). So if you can get all the answers with no learning, was the point getting the answers?

Ben Neely
@benneely.com

Weird fact: in a drop of plasma (0.05 mL) there are ~18 quadrillion albumin molecules, which if you lined them up along their 8 nm dimension you could encircle the earth over 3 times.

Ben Neely
@benneely.com

AI and gaming may be a perfect analogy right now to everything. So I can decomp/recomp anything and play it on any device (let’s play Minecraft ocarina of time on our smart fridge y’all!). But the cost of AI made hardware so scarce we can’t buy devices/consoles anymore.

View full thread
Ben Neely
@benneely.com

Indie devs are empowered too but now they’re stuck with AI-generated competition (steam next fest has like 8k new games this year?!!!). Do they need patrons now? The thing that makes a field/domain/sector accelerate to all dreams are possible almost immediately kills said field.

Ben Neely
@benneely.com

I do think there’s some alternative(s) but I don’t know what they are and the pace of change seems to rapid for sane adaptation.

Ben Neely
@benneely.com

AI and gaming may be a perfect analogy right now to everything. So I can decomp/recomp anything and play it on any device (let’s play Minecraft ocarina of time on our smart fridge y’all!). But the cost of AI made hardware so scarce we can’t buy devices/consoles anymore.

Ben Neely
@benneely.com

Then, companies [will] realized they should stream to solve hardware and game decomp issues. But instead we will get mobile gaming with P2W/microtransactions (except for giants like GTA6). But we can play all those pre-2026 masterpieces anywhere, right?

Ben Neely
@benneely.com

“But it’s AI slop!” /s Actually you had a human editor getting the shorter report to a more succinct (elegant?) proof. Likely means the other 700 get some sort of a treatment too. The bones are good?

Quanta Magazine@quantamagazine.org

OpenAI’s new proof of the quasi-Riemann hypothesis is, as Jared Duker Lichtman of Stanford University put it, “the most significant AI-for-math achievement so far, by quite a margin.” https://www.quantamagazine.org/ai-proved-the-quasi-riemann-hypothesis-what-happens-now-20261019/

quantamagazine.orgAI Proved the Quasi-Riemann Hypothesis. What Happens Now? | Quanta MagazineUnlike other recent AI proofs, this one has been unpacked by human experts. The implications, one mathematician said, are “staggering.”

Ben Neely reposted

Chris Ashwood
@cashwood.proteaglyco.com

We're constantly trying to improve glycomics reproducibility, which is why we're making our NeuGc benchmark sample set commercially available. Whether you're getting into glycomics, or have been doing it for a while, the benchmark sample set enables meaningful quality assurance for your LC-MS.

Ben Neely
@benneely.com

It’s amazing to me that we are cranking into the 100s, and just proposed like 20 more to US HUPO. Haven’t quit yet I guess, Orsburn.

Ben Neely
@benneely.com

Okay, so some "cheating" I have been doing. Pre-2026 (I think?) I was definitely not letting AI write for me, or not write anything important. But now there are things I am cool dumping on it. So I have gone and gotten my writing that I like (lots of samples) and asked LLM to make style guides.

Ben Neely
@benneely.com

One is for more official writing like a readme or internal reports (I still write papers or things I want to think about), and one for paper reviews. On the latter, it is so nice to take random notes as I am reading a paper, then toss them and style guide in and have it "translate" into a review.

Ben Neely
@benneely.com

Btw, I don't actually read these guides. There are a lot of rules in there that were distilled from my writing trends, but I just dump it in and say follow this guide.

Ben Neely
@benneely.com

So apparently Ben @proteomicsnews.bsky.social went off to London to some International Single Cell Mass Spec (iSCMS) conference and did a special LIVE! episode of #THEProteomicsShow, with co-host Melanie Bailey. They talked to Ed Emmott @edemmott.bsky.social about single cells I assume (and more!).

creators.spotify.comEp 113 - LIVE! from iSCMS 2026 - Dr. Edward Emmott by The Proteomics ShowA special live show from the 2026 International Single Cell Mass Spectrometry Conference (iSCMS) in London. Ben Orsburn was joined by special co-host Prof. Melanie Bailey from King’s College London, a...
Ben Neely
@benneely.com

A new math/modeling peptide spectra thing with potential de novo awesomeness from a killer author team (what’s the story on this btw?)… will read on Tuesday for sure. arxiv.org/abs/2610.06282

arxiv.orgdIon: Fragmentation-Based Invariance for Self-Supervised Learning of Tandem Mass SpectraWe introduce a novel invariance for peptide tandem mass spectrometry data, unlocking self-supervised representation learning that improves de novo sequencing of peptides. This invariance exploits the ...
Ben Neely
@benneely.com

(It is taking its sweet time showing up on Apple, will see if it resolves or will investigate. It’s all magic of the interwebs how I got it working.)

Ben Neely
@benneely.com

Not science: just watched the K2 winter summit documentary (apple TV) and there is a telling line about how on summits the white folks would leave the sherpas at camp 3/4, and Nims reversed it. It does seem somewhat good that Edmund didn't ditch Tenzing, and VERY excited to watch this.

The A.V. Club@avclub.com

Genden Phuntsok and Tom Hiddleston climb Everest and a conventional story in Tenzing, which at least highlights its undersung heroes. tinyurl.com/23vfkwv5

Ben Neely
@benneely.com

Do you know someone who knows lotsa cool stuff (macro to molecular) about sweat, ear wax, stomach/bowels, or the placenta? Cause we need them for #THEProteomicsShow. I am striking out too hard on these. Person just needs to be excited to chat, expertise is optional. DM or email me.

Ben Neely reposted

Will Fondrie
@handle.invalid

We released Ptarmigan-1 last week, our structure-free AI model for proteome-wide drug discovery. The model reaches all proteins, including the 40% that have no fixed structure, and can run 3.4 billion compounds against every human protein in a day. 🐦 Try Ptarmigan-1 for free: ptarmigan.app 🧵 (1/8)

ptarmigan.appPtarmigan-1 — Protein–compound predictions by Talus BioPrioritize compounds for experimental testing with Ptarmigan-1. Screen protein targets against compound libraries without requiring a protein structure.

Ben Neely reposted

Michael Lazear
@michaellazear.bsky.social

Have you ever wanted to run a proteomics search engine directly in your browser? No, I haven't either... But now you can! All of your data stays on your local computer, and you can run Sage (with some experimental features 😀) through a nice UI without installing anything. sage.lazear.org/search/