Joe Shockman’s Post

Couldn't have said it better myself “If AI cannot be trusted then AI is effectively useless” https://lnkd.in/gqraDZhR

I would like to take a moment to express my gratitude to some chill VC people I've been talking to. This is a crazy time and little bits of grace shown by awesome people make a whole lot of difference. To everyone out there making this a better place by just being cool to people, I salute you.

s/linkedin/the whole internet/ Everyone is rushing AI junk into everything and it is super irritating. This from someone who actually believes in the promise of AI. I can't imagine how irritating it must be for the users in the middle part of the curve who aren't interested at all till it shows enough promise to overcome the "It worked before, why change it?" barrier

We launched Study Recon on product hunt today. Use it to get an overview of the current state of scientific research on any topic. This is great for scientific researchers but also for students and anyone exploring a new scientific topic. Sign up for free and help me reach my goal of 1M free users! Note: this is a fantastic example our philosophy of fact respecting generative AI. If anyone claims that hallucination free generative AI can't be done, point them at our tool for proof that it can be done right.

I've been blessed with some truly remarkable mentors, friends and leaders. The people I worked with at YouTube were some of the smartest and most capable people you can imagine. They welcomed me, trained me up, challenged me and made me continuously reach for more. I can't properly express how grateful I am to them and the opportunities they've opened for me. I sometimes talk about how good leaders make you more than you were, but it wasn't just the leaders. Almost everyone there had that quality.

I'm fascinated by research to understand what's going on under the hood of language models. I'm following the work at Anthropic closely. https://lnkd.in/gB9bi5ei

This is a key insight (which is to be expected coming from Jeff Hawkins) the human brain is structured into distinct working units. We know from brain damage cases that it's possible to damage your associative cortex rendering the patient incapable of identifying the left side of things. In some ways the LLMs we see to day are only the language center. They lack longterm recall, procedural capability, critical thinking centers and visual centers. Trying to cram a person into an LLM is absurd when you think about it.

Just finished a fantastic meeting with my co-founders at Grounded AI about the future of AI and our place in a world where fact respecting AI is the norm. We look around at all the ways AI is being misused and it is terrifying. But we can also see many cases where proper use of AI could move us forward faster and better. The most egregious misuses are in domains where strict adherence to fact are the most important, such as the domain of Science. False facts making their way into science creates a cascade of error and impedes future science. The recently retracted work on Alzheimer's is a great example https://lnkd.in/gut844mH Thousands of papers cite this work and all of those papers now need to be reviewed, retracted or re-worked and re-submitted. The volume of scientific papers being generated has 10x-ed in the last two years. Most of the increase is from untrustworthy sources, AI generated crap and otherwise fake science. We desperately need to address that or risk devaluing science itself. But this is also a bigger issue for the future of AI and the trustworthiness of the information we consume. If we can get this right today we usher in a beautiful new world where AI accelerates and assists us. If we get it wrong we will create a hell of our own making where nothing can be verified and everything is bullshit. We are early enough that the work we do today can influence which future unfolds. I can tell you one thing. I know which way my work (and my company's work) is pointing us.