I honestly think FastGPT is the best implementation of AI w/search, and is extremely versatile/useful across domains (granted I don't code). I think it's the same thing as the Quick Answer feature from their standard search.
In my daily work as an MD it's become my reflexive go-to for looking up answers to specific to general, easy to complex clinical questions. I use it far more frequently than UpToDate (which is no less than the holy book of medicine), more than PubMed/Google Scholar searches, and definitely more than a basic web search (Google, only b/c it's a hassle to log in to Kagi every session at work).
Maybe 1 time out of 10 it won't give a correct or meaningful answer (in which case my prompt needs to be refined, or is just not suited for this kind of tool). But apart from that it will give me exactly what I need, because it uses Kagi search to source its answers. Kagi search does a decent job bringing to the top relevant journal articles (which in turn may mention other articles, adding indirectly to the trove of sources FastGPT pulls its final answer from). It shows the 5 search results it referenced at the bottom of the page, so more often than not if I don't get my answer in the direct summary, I have very relevant sources to read through.
I also don't think you need a Kagi account to use it.
Sourcing information from better search results. I cancelled my Perplexity Pro since for any use case I had for it, I would instead use Kagi FastGPT. I tried Assistants (beta) but I didn't think it was anything special, didn't really see a way to integrate it as a daily tool, and ultimately FastGPT gave me the best answer, even better than their gpt4o and Claude Opus/Haiku based assistants.
I really love this app/concept. I'm curious from your time working on the app what applications/use cases beyond note-taking you might have uncovered, that might not be obvious at first glance? (e.g. If studying a topic, creating a note of questions to quiz yourself from time to time, with the answers likely being found in semantic search.)
There are scores of medications necessary for chronic conditions that we give to patients without intention of stopping that have not been around for a lifetime of testing for “long-term safety.” If your cardiologist prescribes you an SGLT2-i for heart failure because we have studies that show it has significant mortality benefit, you probably won't bat an eye before taking it---yet they have only been FDA approved since 2013. (As an aside, people also don't generally rag on heart/kidney/liver failure patients for poor lifestyle habits that may have contributed or continue to contribute to their disease, as they do for patients with obesity.) GLP-1 receptor agonists like Ozempic have been on the market since 2005. Obesity is among the most prevalent, morbid diseases that itself accounts for 20% of US health expenditure; it will lead to diabetes, hypertension, OSA, heart failure, arthritis, depression, cirrhosis, even cancer and then some. Our scientific and clinical understanding of obesity is evolving beyond it being a self-imposed and self-treatable lifestyle choice, but as a complex neurohormonal disease; one that is chronic like hypertension, and that must be treated chronically (like hypertension) as well. As doctors, counseling patients to diet and exercise (even offering nutritionist referrals) is the equivalent of sitting idly by as patient after patient rotates through our clinic having tried diet/exercise for years all the while developing obesity-related comorbidities.
i would love if years later it turned out he didn't just have "nobelitis" and was right in some respects. but this is one small pilot study and the larger body of literature around it shows mixed results. not saying it should be disregarded, but there are several reasons why it's not exciting and it shouldn't be mistaken for a anything more than a pilot trial to evaluate safety.
i'm sure there's a lot more to criticize than i can see, but:
1) study's primary outcome of increased urine output is not necessarily clinically meaningful---urine output (UOP) is selected because in septic shock (sepsis + low blood pressure), organs slowly shut down with the kidneys typically being the first to go. thus, UOP can be informative as a measure of renal perfusion and ultimately end-organ damage. however, the control arm received D5 fluids [water? saline?] alone, whereas intervention group received D5 fluids + NaAscorbate, thus ultimately receiving more solute. you would expect when given a more osmolar solution, you would see increased diuresis (solute will draw in water in the kidney's tubules, thus more urine output). additionally, vitamin c itself is renally excreted, thus if the kidneys are working to excrete high doses of NaAscorbate, the tubular cells could alter their usual reabsorption and secretion patterns, potentially affecting overall fluid balance.
2) as above this is a small pilot study; however, the authors (laudably) include their power calculations in which they used an effect size of 900cc UOP. the difference in mean UOP between the groups was 891cc's, so it's dubiously powered (even if we ignore consideration (1)). accordingly, their 95% confidence interval itself crosses 0 meaning no difference in effect was detected in urine output (thus their p values was >0.05). you have to wonder what statistical games might have been played to p-hack when results come so close.
ultimately why this isn't news is that the 'next phase of human trials' have already been conducted on vitamin c in sepsis.
there's a few other things: mean baseline CRP was notably much higher in the intervention group, and it makes me wonder if this affects the effect size considering that anyone in sepsis might benefit from fluids (though they received so little overall).