That issue is different, when web tools were added to gpt4o it would fetch the site, and basically copy paste the text into the answer body. So, you were able to read the content of the site without the site getting the ad impressions. Now the system prompts put a very tight word limit - 25? - on quotes from sites the model visits
I’ll save you a click: ‘Llms can’t perform direct calorimetry through a photo of a meal. Llms can’t even perform basic atomic spectroscopy’ in other news…
I suspect this is part of the reason why gemini 3.1 pro is insanely good on AiStudio and pretty bad on the gemini app. I have thousands of small videos to convert to detailed descriptions and I'm using a super detailed system prompt. It works perfect either via api or Aistudio. I tried doing a gem on the gemini app using the same prompt as the gem instructions and I just can't get the same results. So, the issue might be not just the rlhl but also the massive system prompts injected on the app interface
Many years ago I had a LG CRT that often would turn off in the exact moment I entered the bedroom. The monitor was configured to go to sleep after 20 or so minutes of inactivity, so it was supposed to turn off on its own. And I always assumed it was a case of only noticing it when it did happen (like when you buy a car and suddenly start seeing the same model everywhere)... But it always seemed uncannily frequent... Now I wonder if I might have somehow disturbed the electric field each time
My wife has a very serious health issue, that has caused more suffering then words could describe. o1-preview was the first ai that actually proved useful. From there on, each improvement on ai caused an incremental improvement in her situation. Even recently we were able to pinpoint exactly what was causing her flare, and solve the situation the same day, just by prompting a claude opus conversation where i’ve shared all her health notes. But if i weren’t a data freak and haven’t been collecting data about her issues (what she does/takes and how she feels) for so long i dont think we would had been able to get this far. So i think ai appeals to people with problems that can be solved by finding patterns in data. People that say ai makes mistakes don’t understand that the power is in finding patterns, not in finding THE right answer. You need to prompt from that prespective
Wasn't Antrophic previously offering double the token usage outside busy hours? Now they are counting tokens back at normal rate. But yeah, it's not good. I use codex because claude insists in peaking at and messing with folders and file outside its work area though
I think a cool use would be as an harness for agents or automation workflows, to simplify the page/remove clutter to facilitate navigation. I.e. have the agent send commands to your tool
Does asking it to think step by step, or character by character, improves the answer? It might be a tokenization+unawareness of its own tokenization shortcomings
As someone who lived through the blackout it was wild. I felt back into the pre-internet, pre-smartphone era. It was pretty cool actually. The rumor mill spread so fast that Within hours the official word on the street was that we were getting hacked by a foreign military and people were joking that we had nothing of interest to be conquered xD
I have used chatgpt 5.2 thinking for health, gemini hallucinates a lot, specially with dna analysis. Never tried using the new claude even though i have access through antigravity. Might give it a try. Do you have any tips on how to approach it for health ‘analytical power’?