#161

Yes, prompt-engineering. I just discovered that it’s a thriving science. Apparently, it is not just the model you use, but rather the way you design your prompt that can make you hit the target.
Also, it is useful to use a master-prompt, a document where the rules and preferences and instructions are clearly stated, and submit it to ChatGPT as an attachment or to paste it to the beginning of the prompt.
Models like Google’s Gemini have settings where you can build something similar to a master prompt and edit it. Maybe that’s a fetaure of ChatGPT as well.
There are so many other novel concepts, like canvassing that I need to grasp thoroughly.

1 Like
#163

I went on asking lifestyle interventions to inhibit upstream signaling, interesting answers, intuitive aspects we already knew.
For example, the Rosedale diet, low protein + low carbs, brutally hits Hits amino acid sensing, insulin/AKT, and AMPK–TSC (if we also exercise), causing inhibition of mTOR by natural means.
Of course, this would also sends a strong catabolic signal with loss of muscle mass if done chronically.

#164

A diet high in fat is also high in saturated fat and will kill you faster than cancer from high mTOR activity will.

1 Like
#165

I’m trying now to share through a link my A&Q with chatGPT5, deserves to be read to understand the power of prompting (and the upstream signaling in mTOR). According to what probably thought DAvid Sabatini, levels of probability included!.

1 Like
#166

Why is perplexity.ai garbage now? It’s making up words and when you ask follow ups it forget the previous statement like they cleared the context window. Did they run out of VC money and have to cut costs making it unusable? It’s giving terrible first impressions to new users (when logged in they probably don’t do these cost cutting measures).

It was better like 2 years ago lmao.

1 Like
#167

Derya Unutmaz (Jackson Lab researcher) shares some of his success with using GPT-5-thinking to analyze large datasets:

https://x.com/DeryaTR_/status/1957983877114339465#m

I am now sharing part two of this amazing data analysis performed by the GPT-5 thinking model. My goals are twofold: (1) to show the power of this model in analyzing complex, large biological datasets, which can help scientists studying similar conditions to perform these sort of analysis to develop novel hypotheses to test; and (2) to give patients hope that, using such detailed datasets, using AI we will be able to develop personalized treatments or identify targets at speeds that were unimaginable even a few years ago!

There are some interesting findings in his tweet. It will take me a while to digest it (using GPT-5-thinking to answer questions).

1 Like