idk, I would say that at least in healthcare, most people are sufficiently terrified of the consequences of HIPAA violations that they won't access records they aren't supposed to. In this case, what works is a combination of serious, career-ending consequences and the knowledge that compliance departments conduct regular audits of everyone's access logs.
Remember that you still need to include an instruction to produce a CSV to get the prompt into the right context to generate a CSV that makes sense. Otherwise, you may get output that is technically in the CSV format but doesn't make any sense because the model was actually trying to write a paragraph response and the token sampler just selected really low-probability tokens that the model didn't really want to say.
I agree with you, but I wonder what the incentive is for developers to use resources responsibly. Is it just that users will uninstall apps that drain their battery? Or are there penalties imposed by the OS on developers who refuse to play nicely?
> fixing spelling and grammatical errors should involve humans in the process, so you really can't automate this
This is an interesting observation to me. I would have expected that, since LLMs evolved from autocomplete/autocorrect algorithms, correcting spelling mistakes would be one of their strong suits. Do you have examples of cases where they fail?
I understand education and healthcare, but how do trade skills and housing require human interaction? For housing especially, a lot of foreign/remote investors can own houses and just collect rent checks from tenants in a pretty hand-off manner. Housing is supply-limited, sure, and heavily regulated, sure, but I don't think it requires human interaction, and certainly not to the scale of education or healthcare.
Is it decided then that screenshots are better input for LLMs than HTML, or is that still an active area of investigation? I see that y'all elected for a mostly screenshot-based approach here, wondering if that was based on evidence or just a working theory.
Maybe it's better to build specific applications with OpenAI's products (e.g., using the API to automate a specific business process in your company) than to build a company offering a generic additional functionality like a UI or fine-tuning or custom prompts. Or at least, that seems like the target audience that OpenAI wants to sell their product to.
> The real value of this process is looking at your data and doing careful analysis. Even though an AI judge can be a helpful tool, going through this process is what drives results. I would go as far as saying that creating a LLM judge is a nice “hack” I use to trick people into carefully looking at their data!
Interesting conclusion. One of the reasons I like programming is that in order to automate a process using traditional software, you have to really understand the process and break it down into individual lines of code. I suppose the same is true for automating processes with LLMs; you still have to really understand the process and break it down into individual instructions for your prompt.
As a medical student, perhaps I can give some recommendations on the best free resources to learn medicine. If you like YouTube videos, Ninja Nerd has a great channel for learning the foundations. If you like textbooks, I used Guyton for physiology and Harrison's is probably the standard for clinical medicine. If you just want to look up how to treat a specific condition, look up "<condition> clinical guidelines". You'll always be missing the additional knowledge that comes from years of experience, but there's no harm in increasing your knowledge as long as you maintain the humility to remember that your knowledge is incomplete.
It's similar to how I learned software development as a hobbyist, so I understand a little about headlines like OpenAI switching from Next to Remix, but at a deeper level, I don't really understand what it's like running Next.js at the scale of MAUs. But it's still worth learning so that I have a little more understanding about the world around me.
As an armchair economist, this is my personal favorite theory. With one of the most obese populations in the world, I think it's obvious that we'll also be the most sick and we'll have to spend a lot to try to dig ourselves out.
Okay, but I wouldn't be convinced until I've counted up to 756,912 myself and seen that there are no more leaves left after that. Isn't it supposed to be less computationally intensive to verify the proof than to complete the original calculation?
I tried learning about zero knowledge proofs during the crypto craze, but I never understood the basic idea of how it was supposed to work, much less the math involved. I'd appreciate an ELI5 from anyone who has one.
What's wrong with the simple solution, i.e., understanding when the algorithms apply and when they don't? For example, if an RCT shows that in-hospital initiation of all 4 of the heart failure GDMT's leads to reduced mortality compared to delayed initiation of GDMT, then you also have to learn that 97% of patients in the study had a baseline SBP of at least 97. Therefore, maybe you shouldn't start 4 BP-lowering meds on your HF patient with a baseline SBP of 86. If people learn when to apply rules, then I think you don't need humanities education and you don't need a system-wide reset either.
> the cost of shelter accounts for 70% of all inflation for the last twelve months
Technically, this should read "70% of inflation in all items less food and energy". And since enrgy costs are actually down for the past year, this basically says that rent increases are the majority cause of inflation except maybe food, but that's not clear from the BLS data linked.
What I'm interested to see is how this will spur additional divestment from Chinese supply chains in Western countries. Now that Israel has set a precedent, Western governments will likely be even more hesitant to use any goods that touched Chinese territory during production.
I feel like this is the opposite of what LLMs are useful for. I like using LLMs to summarize and get an immediate answer to a specific question, like the AI-generated summary in Google Search. In that case, the increase in convenience outweighs the decrease in reliability. But if I wanted to read a full article about a topic, I would no longer be concerned about convenience, so I would look for a more reliable source than an LLM.
Yes, but the CEO was trying to take those profits for herself by biding to take the company private, which the Board of Directors refused to put up with
I actually think we should be decreasing demand rather than increasing supply. Not every job needs a college degree, and allowing more people to avoid college altogether would naturally drive prices down from the demand side. The hard part is that I don't know how to make employers accept non-degreed applicants, outside of having Peter Thiel shout at them to do so.
My understanding of a free education is one that is 100% paid for by the government, since someone still has to pay the professor's salary. In that case, you still have an issue of an industry that can demand increased prices (from the government, of course) if university enrollment suddenly increases (and I think it would if students didnt have to pay tuition). So I think you still have to figure out how to handle the market forces, but you're right that shifting the burden from the students to the government lets people start off their lives with less debt, which is a benefit of its own.
I mean, a lot of trades do have standardized tests. The LSAT for law school, the MCAT for medical school, and the CPA exams for accounting come to mind. I don't know why computer science hasn't organized in this manner, maybe because it's a younger field or maybe because there are simply too many people and employers don't care enough about having certified employees.
Interested to hear what you mean when you say "turn clinical researchers into real scientists." Are you thinking of expanding the MSTP? Or just better statistical literacy for physicians? Or something else entirely?
I haven't used Character.ai personally, but assuming the user base is the same as the Silly Tavern userbase, and assuming that the Silly Tavern user base is accurately represented by the subreddit, then it's like 95% ERP