Anthropic AI tool sparks selloff from software to broader market
bloomberg.com
bloomberg.com
How could it possibly keep up with LLM based search?
I believe frontier labs have no option but to go into verticals (because models are getting commoditized and capability overhang is real and hard to overcome at scale), however, they can only go into so many verticals.
Interesting. Why wouldn't an LLM based search provide the same thing? Just ask it to "use only trusted sources".
Oh so they are not just helping in search but also in curating data.
> They've got hundreds of thousands of physicians asking millions of questions everyday. None of the labs have this sort of data coming in or this sort of focus on such a valuable niche
I don't take this too seriously because lots of physicians use ChatGPT already.
Is that sarcasm?
I think what you'll end up is a response that still relies on whatever random sources it likes, but it'll just attribute it to the "trusted sources" you asked for.
Please read what I have written clearly instead of assuming the most absurd interpretation.
You started off by asking a question, and people are responding. Please, instead of assuming that everyone else is missing something, perhaps consider that you are.
Here’s what I mean: LLMs can absolutely be directed to just search for trustable sources. You can do this yourself - ask ChatGPT a question and ask it to use sources from trustworthy journals. Come up with your own rubric maybe. It will comply.
Now, do you disagree that ChatGPT can do this much? If you do, it’s almost trivially disprovable.
One of the posters said that hallucination is a problem but if you’ve used ChatGPT for search, you would know that it’s not. It’s grounding on the results anyway a worst case the physician is going to read the sources. So what’s hallucination got to do here?
The poster also asked a question “can you ask it to not hallucinate”. The answer is obviously no! But that was never my implication. I simply said you can ask it to use higher quality sources.
Since you’ve said in asserting BS, I’m asking you politely to show me exactly what part of what I said constitutes as BS with the context I have given.
More seriously, the concept of trust is extremely lossy. The LLM is gonna lean in one direction that may or may not be correct. At the extreme, it wound likely refute a new discovery that went against what we currently know. In a more realistic version, certain AIs are more pro Zionist than others.
The thing is, LLMs are quite good at search and probably way way more strong that whatever RAG setup this company has. What failure mode are you looking at from a search perspective? Will ChatGPT just end up providing random links?
Yet now I provide an example of a very recent, big, very obvious, very prominent security explosion and now I am "grasping at the latest straw".
Ok man.
I’m guessing you’re not even aware of what OpenEvidence is, nor are you aware that every doctor you know uses it.
There is trust and then there is accountability.
At the end of the day, a business/practice needs to hold someone/entity accountable. Until the day we can hold an LLM accountable we need businesses like OpenEvidence and Harvey. Not to say Anthropic/OpenAI/Google cannot do this but there is more to this business than grounding LLMs and finding relevant answers.
And how does the LLM know which specific sources to ground itself to?
This is pure LLM brain rot. You can’t “just ask” an LLM to be more reliable.
For example, only 7% of pharmaceutical research is publicly accessible without paying. See https://pmc.ncbi.nlm.nih.gov/articles/PMC7048123/
Edit: seems like it is ~10M USD.
We turned that into a proper, ready-for-use-in-AI dataset and contributed it to the mainstream AI community under the name OpenDebateEvidence. Presented at NeurIPS 2024 Dataset and Benchmark track.
(My wife's a hospital doctor & author and introduced me to it; other family in other countries.)
See this. I use OpenEvidence. It has access to full text from some of the major medical journals. But generalist models seem to outperform it. Not sure what is going on there.
The assumption is that Claude has access to a stream of fresh, currated data. Building that would be a different focus for Anthropic. Plus Thomson Reuters could build an integration. Not totally convinced that is a major threat yet.
Again, unless Anthropic are taking on liability for their legal tools, this is not going to impact TR.
That being said, there probably is a potential company here that's gonna be built soon/is currently being built, but it definitely won't just be a wrapper around Claude as the recall will be way too low for these systems unaided.
You can be a huge, profitable data-only company... but it's likely going to be smaller than a data+interface company. And so, shareholder value will follow accordingly.
If AI tooling starts to seriously chip away at those foundations then it puts a large chunk of their business at risk.
Capital also won't be rewarded to people who don't have privileged/proprietary access to a market or non-public data or methods. Just being a good engineer with Claude Code isn't enough.
No, it will just lead to the end of the Basic CRUD+forms software engineer, as nobody will pay anyone just for doing that.
The world is relatively satisfied with "software products". Software - mostly LLM authored - will be just an enabler for other solutions in the real world.
> The world is relatively satisfied with "software products".
you can delete all websites except Tiktok, Youtube and PH, and 90% of the internet users wouldnt even notice something is wrong on the internet. We dont even need LLMs, if we can learn to live without terrible products.
I'm like: oh that's it, a bunch of skills files?
So the value of a skill file is that it tells the model how to format its response for use within the software environment surrounding the model.
With programming, it's mostly about how to tell it to use some API.
But all the model can do is reply some text, and the actual work needs to be done by the software(the agent harness) which needs to parse the model response and translate it into actual work.
My point is there is no magic: the model just reads the skill file and then uses that as a template for a textual response, which is then parsed and processed by traditional software.
So in terms of legal skills, a stand-alone skill like the contract review skill at https://github.com/anthropics/knowledge-work-plugins/blob/ma... is basically useless.
Yes, the model will read it and it will influence its response, but without some extensive software harness around the model to give it data for context and and so on: totaly useless.
Why? Because garbage in is garbage out.
So telling the model to review a contract and pay attention to "Whether indemnification is mutual or unilateral" will result is some response from the model, but without additional data it will be at the same level as what you can get from a google search.
The effect on established companies is exactly zero.
Now, having an in-house skills and proprietary software around the model to integrate it into your system, that would be valuable indeed, but not something an AI lab can replicate without building the whole company from scratch.
If that happens, some software companies will struggle to find funding and collapse, and people who might consider starting a software company will do something else, too.
Ultimately that could mean less competition for the same pot of money.
I wonder.
Something seems quite off. Am I the only one?
I might be not across the detail, but to me the legal plugin seems like it’s mostly adding some skills (prompts) that are fairly basic that any technically minded people could do, and is not enough of an improvement for completely non technical people to use.