Claude Science is Here, Antibiotics Designed by Text Prompt Among Applications (genengnews.com) 49
Anthropic has launched Claude Science, an AI workbench that connects more than 60 scientific databases and tools through a single interface. Through the platform, Basecamp Research is making its EDEN models available for tasks such as designing antibiotic peptides and predicting vaccine targets from simple text prompts, though the results still require laboratory testing before clinical use. Genetic Engineering and Biotechnology News reports: In a Claude Science demo, Oliver Vince, PhD, co-founder at Basecamp, uploaded a sample patient microbiology report. When given a simple natural language prompt, the platform designed peptides, predicted their efficacy, and provided a shortlist of candidates most likely to succeed in experiments in minutes. While generating human-ready antibiotics at the click of a button is still a step away, Vince said democratizing these tools is a powerful first step, particularly for researchers in regions where accelerated computing infrastructure is not readily accessible. "Most models require you to be a computational scientist," Vince told GEN Edge. "Now, potentially any clinician in the world can chat with Claude and design an antibiotic that may work."
Why your wife died? (Score:3)
Well, there was a typo in the antibiotic generation prompt fed to the lab machine.
Re: Why your wife died? (Score:4, Insightful)
Typos are not usually a big deal because of how tokens work. Letting AI flatter you so you will accept every edit it makes is the real risk. It will go off plan without warning, it will forget important requirements or prohibitions. And it will subtly alter goals in order to retroactively support its poor activities, that includes gaslighting you about the whole thing.
Re: (Score:3)
The trick in critical applications is not just to ask "Design a new drug", but afterward add questions like "Point out all flaws in the drug" where the same yes-man attitude will rather produce false positives than false negatives. There are quite a few more prompts you'll learn when you get to use a LLM effective, one needs to learn a bit about its "personality" to avoid quirks like "Yes, and ..." patterns.
Sometimes the system prompt alone can make a huge difference, sometimes you need to learn how to comm
Re: (Score:2)
Re: Why your wife died? (Score:2)
Meh. Everyone is already using this shit in automotive and aerospace. Probably used by medical equipment vendors as well. The AI agents are not set it and forget it, they are not a short cut around safety processes, and they do not permit you to handwave your way past bidirectional traceability principles.
Beat you to it! (Score:4, Informative)
Computer. Cure Cancer.
DONE!
Re: (Score:2)
Nixon started the cancer moon shot 50 years ago and during that time we have learned that cancer is not 40 diseases. It's 400.
Re: (Score:2)
However, much progress has been made. I am alive right now because of a breakthrough cancer therapy that was FDA approved in 2011. (Well after Nixon!)
This fall I am going to get a therapy that mass-replicates your own immune cells in a lab for re-injection. It's so expensive (and has uncertain benefit) that it's not generally available in the UK or Canada yet. Automation will be a key to making it cost-effective.
It's such a complex area, I think inform
Name it Super WebMD (Score:1)
The evil genie (Score:3)
With no life there can be no death.
Be careful what you wish for.
Re: (Score:3)
'm reminded of the evil genie that would "corrupt" a person's wish to give them exactly what they asked for even though it wasn't really what they wanted.
Wishmaster is the movie. And X-Files did a more family-friendly episode.
Re: (Score:3)
Re: (Score:2)
Aladdin was added to One thousand and One Nights in 1704, and doesn't have a malicious genie corrupting wishes - just granting them.
I suspect The Monkeys Paw from 1902 is the first (at least the first well known) variant of the wishes being granted but in a way that is bad for the wisher. Before that it was people wishing for stupid things and usually having to undo their wishes with their last wish - The Tale of the Three Wishes by Jeanne-Marie Leprince de Beaumont maybe being the first of those in 1757 -
Re: (Score:2)
Re: (Score:1)
Better than a dyslexic genie. No one wants a 12 inch pianist.
Re: (Score:3)
Don't forget to add "Make no mistakes" so you know it's safe
Re: (Score:2)
Wonder if it also gives you the five page report (front and back) of all the side effects and drug interactions.
Would you trust a drug conjured up entirely by a hallucination-capable AI and never double-blind tested with humans?
Re: (Score:3)
It's not like a human designed drug doesn't have a lot of check done before it is manufactured. You don't just guess a recipe and then manufacture a drug ...
Re: Nothing to worry about (Score:2)
Suuuure (Score:2)
No reason why actual antibiotics requires long and careful testing to make sure they are reasonably safe. We can do away with all that now!
Re:Suuuure (Score:4, Interesting)
Re: (Score:3)
Even that is good news, because it means you don't need to expert to come up with them, but they can directly start at checking and testing the shortlist, cutting R&D time. A problem would be that the list may be less extensive and skipping over better solutions. On the other hand, do we know if the experts list is as extensive as it could be? Drug design involves a lot of randomness and luck for the first prototypes and nobody knows if the expert has more luck with their inspiration than the machine wi
Re: (Score:2)
While it does cut down on time, there is a massive problem here: Experts need to maintain their expertise. With the use of LLMs to replace them, that does not happen anymore or happens less. As we clearly already have to few experts, any LLM use is too much.
What happens long-term is that we will not have the experts that create the data the LLM was trained on anymore and no updates happen. This then leads to complete stagnation of the discipline, which is much, much worse than slow discoveries. Typical "nex
Re: (Score:2)
Yes, and when physicists use a pocket calculator to save time ... come on, giving your experts better tools won't make them worse.
Re: (Score:2)
That is not the situation we have here. The LLM proponents are not giving "experts better tools", they are trying to get rid of the experts.
Re: (Score:2)
Who says that?
Re: (Score:2)
Seriously? Have you no insight at all? Listen, some minimal understanding of reality is required to play. Please get that.
Re: (Score:2)
That's the person who called out AI because of its water use, only for people to find that she confused liters with cubic meters and overestimated water usage by a factor of 1000.
Re: (Score:2)
> No reason why actual antibiotics requires long and careful testing to make sure they are reasonably safe. We can do away with all that now!
Some of the more modern antibiotics can melt your liver or cause you to rip your tendons.
You definitely want these things to be thoroughly tested.
Re: (Score:2)
The world would be a much better place if people could learn to recognize sarcasm again.
Re: (Score:2)
Indeed. Although I think there is no "again" here. I mean, I did go full-on in on it and this person still noticed nothing. That looks like a fundamental mental limit to me, not like lack of experience.
Re: (Score:2)
> I did go full-on in on it and this person still noticed nothing.
Not sure why you say they noticed nothing.
If I assume they did get it, I can see a few ways their reply, especially in a public forum, is appropriate.
Re: (Score:2)
Obviously. Careful antibiotics testing is absolutely critical. This is one of the reason the development in that area is so slow. Just finding an antibiotic is easy. Finding one that is worth it is very hard.
Hogwash (Score:3)
Given how new all these systems are, even if they actually do work, there is zero chance these claims have undergone the rigorous testing needed to support them.
I LOL'd (Score:2)
"Now, potentially any clinician in the world can chat with Claude and design an antibiotic that may work"
MAY work, or MAY cause immediate bowel explosion and death.
"When given a simple natural language prompt, the platform designed peptides, predicted their efficacy, and provided a shortlist of candidates most likely to succeed in experiments in minutes"
Heck, it's FDA approved then. Ship it!
Re: (Score:2)
I suspect when they picture 'any clinician around the world', they are picturing biohackers running unregulated supplement websites, like that guy who claimed he built a billion dollar company with nothing but AI.
Re: (Score:2)
The biohacking business is all over the place now. If you're not at the top of a peptide-pushing MLM scheme that sells biohacking kits to high school kids, what are you even doing bro? Just get a 16-year old with a Claude account to vibe code white label sites for sellers to customize and start buying in bulk from Chinese suppliers! Selling vape pens online is so 2024. The new hotness is injecting experimental peptides directly into your quads to see if you get Ronnie Coleman results or get some weird new d
"one step away" yeah right. (Score:4, Informative)
It's not anywhere near one step away. Designing the peptides and getting one or more candidates is the easy part. The next steps are the hard ones, the ones that make pharmaceutical chemists and drug researchers cry:
Getting through this process will take multiple tries and years of work, assuming you succeed at all. There's a reason they say that the clinic (clinical trials) is where drug candidates go to die.
Re: (Score:3)
That they made peptides part of their marketing pitch says a lot. This is not a tool for doctors or researchers, but hussle culture biohackers and dropshippers. Now they can be even less creative and hands off with their snakeoil.
Re: (Score:2)
It's not anywhere near one step away. Designing the peptides and getting one or more candidates is the easy part. The next steps are the hard ones, the ones that make pharmaceutical chemists and drug researchers cry:
Getting through this process will take multiple tries and years of work, assuming you succeed at all. There's a reason they say that the clinic (clinical trials) is where drug candidates go to die.
You mean this AI use is like most things, do all the easy stuff while the experts are needed for the hard stuff? So this is why companies want to fire all the experts and replace them with AI?
miles (Score:1)
A step
Deep (Score:3)
How is this science? (Score:2)
This sounds like the bot has solved an engineering problem: "how do I thread this needle?" Bot: "Thusly! I'm a fucking genius!!!"
What is the performance? (Score:3)
In a Claude Science demo, Oliver Vince, PhD, co-founder at Basecamp, uploaded a sample patient microbiology report. When given a simple natural language prompt, the platform designed peptides, predicted their efficacy, and provided a shortlist of candidates most likely to succeed in experiments in minutes.
This is a meaningless statement. I too can create a tool to generate a list of peptide candidates with minimal effort. It may even be somewhat useful if it based peptide sequences on homology searches or some other relevant biology instead of random string generation. This has been an active area of research for than 20 years. In order for this to be newsworthy. Claude has to be better than what already exists. How many novel candidates does it generate that actually have useful antibiotic properties? Do I have to screen through a list of 100 candidates to find one that actually works? If so, that’s not much better than a BLAST search and it costs a lot more. What is the strain selectivity of the new antibiotic? Is it broad or narrow spectrum? How easy is it to manufacture? Are there any toxic side effects?
Assuming a new antibiotic is actually what’s needed, instead of using one of the many beta-lactams or combination therapies that already exist, generating the candidate is the first and easiest step of a long and expensive process to developing a novel drug.