I mean, how about neither?
1 Like
That guy is pretty batshit. You should read his blog to gain an understanding of just how warped his views are.
Him and Thiel are two very problematic people.
This is generally what happens when you get two techno-geeks with very limited understanding of the social sciences (very low emotional IQ as well) trying to project their dysfunctional views onto society. They also live in a very narrow bubble that further amplifies their views.
These are not the type of people you want influencing policy decisions.
1 Like
Our new tech just started on Thursday. She needs to take 25,000 pictures which will then be individually annotated by two peoplefor a project Iām involved in to develop a task specific AI.
He sounds like an awful person.
Silicon valley trump
This makes AI sound like a bit of a house of cardsā¦
https://www.nytimes.com/2025/11/08/business/dealbook/debt-has-entered-the-ai-boom.html?unlocked_article_code=1.z08.vTF2.RPA8fvx7juuI&smid=nytcore-android-share
Consumers are spending $12 billion/year on AI and thereās a need to spend $7 trillion on data centers by 2030 and to get their companies are using exotic financial instruments. The math isnāt mathing for me.
Why doesnāt Thiel just pick a religion thatās more tolerant of homosexuality? You can see the mental torture on his face every time he appears in public.
We may find a better way to get AI with less raw input processing power.
Fixes the āfattiesā problem, too! Win-win!
Was it out of a window?
11 Likes
Iām trying to get the compliance department of an insurance entity to agree that public AI services arenāt compliant. Iāve got that far. Now I need them to take two more steps; one is to push out that fact to everyone theyāre associated with (thou shalt not use public AI) and thenā¦declare that there is a compliant solution and itās hosted by the one and only space lobster.
Still working on steps 2 and 3 but I have an inside player whoās assisting.
So yesterday I spent a few minutes on an AI directing it to compare various Canadian insurance regulations to what the Ai companies are offering, and provide specific regulation paragraphs that are not compliant. And then contrast that with what Iām offering and show it is compliant.
About 15 minutes of work and I have a working version of a white paper. Now I have to read it, tweak it, and compare it to the actual regulations to make sure itās not hallucinating. But thatās a couple hours of work, not a couple of weeks.
Iām also looking at required hardware. Itās a nightmare of lack of info.
Thereās like almost no way to estimate, given an assumed load on an AI package, what type of hardware is sufficient. Do I need a $10K server? Or do I need a $150K server? That kind of drives the price of the service, hello? And if I spend $10K on a server and later find out I need $100K server, thatās a whole other problem. and on top of everything, the various models have different requirements. And then each model has different versions which use more or less hardware, but produce better or worse results. Geesh.
We tried a few things. We have amazon instances set up, and Iāve thrown a bunch of simultaneous prompts at it. But we tried to do a bigger model, which requires a lot heavier hardware. We got approved by amazon for an instance with the bigger hardware, tried to spin up an instance and it came back with āno available machines in your locationā.
Man, Itās gonna be a run a jump and a wish I think when it comes to picking hardware.
1 Like
So youāre hosting a copy of an existing LLM with no communication back to the LLM creator?
I am curious about how Canada deals with data protection in the context of using LLMs.
You set up an AI agent online via a licensed AI vendor (to hoover up personal and financial data I assume), this then gets processed in the background via the LLM.
Does the LLM creator still own the right to use that data further?
correct.
and gonna install more. I am going to put a copy on the machine with my client database. when I hang up the phone, a script will take the recording, turn it to a text transcript and attach it to the client record. then the local ai will parse the text and create a summary that also gets attached. then it will parse it a second time to check for any tasks discussed, and inject a task in my calendar. thatās going to save me a ton of time and be more compliant than what Iām using.
Yep, they are explicitly using prompts for training. thus the compliance breach. or one of them anyway.
1 Like
āDear GPT - No to snappy, punchy etc. Yes to STFU if I publish in a major national newspaperā.