Jamie Tidman
@jamietidman.bsky.social
CTO at Japeto.ai - currently working on building secure language AI products for healthcare and local government.
Alexa powered by Amazon Titan sounds even less useful than it currently is. Titan is by far the worst commercial LLM I've used.
February 5, 2025 at 9:30 PM
Alexa powered by Amazon Titan sounds even less useful than it currently is. Titan is by far the worst commercial LLM I've used.
There's a bit in a Silicon Valley where they give admin access to an Adderall-fueled teenage coder who destroys the entire system, and I feel like that's about to play out on a massive scale
February 4, 2025 at 4:10 PM
There's a bit in a Silicon Valley where they give admin access to an Adderall-fueled teenage coder who destroys the entire system, and I feel like that's about to play out on a massive scale
This still makes sense canonically given this version of Lorca was from the evil timeline.
Every car is a Cybertruck in the mirror universe
Every car is a Cybertruck in the mirror universe
January 29, 2025 at 2:23 PM
This still makes sense canonically given this version of Lorca was from the evil timeline.
Every car is a Cybertruck in the mirror universe
Every car is a Cybertruck in the mirror universe
I agree with your last point wholeheartedly! My point is that Deepseek is a vindication of LLMs as a technology, not a threat or a sign of failure.
January 27, 2025 at 11:01 PM
I agree with your last point wholeheartedly! My point is that Deepseek is a vindication of LLMs as a technology, not a threat or a sign of failure.
Yes, it is unfair given the advances are promotional to the investment. Deepseek makes that investment more rational, not less.
January 27, 2025 at 10:50 PM
Yes, it is unfair given the advances are promotional to the investment. Deepseek makes that investment more rational, not less.
Fair enough. I’d say given the advances in AI over the last 3 years anything short of ASI is apparently going to be defined as “not enough progress”!
January 27, 2025 at 10:48 PM
Fair enough. I’d say given the advances in AI over the last 3 years anything short of ASI is apparently going to be defined as “not enough progress”!
“Little to no progress”!!
January 27, 2025 at 10:43 PM
“Little to no progress”!!
An extremely significant development in AI efficiency = overhyped?
That doesn't make sense to me.
That doesn't make sense to me.
January 27, 2025 at 8:35 PM
An extremely significant development in AI efficiency = overhyped?
That doesn't make sense to me.
That doesn't make sense to me.
I can see a market for this for some exasperated CISO who fields so many dumb cybersecurity questions that he says FINE WE’RE STORING ALL OUR DATA ON THE MOON
January 25, 2025 at 1:02 PM
I can see a market for this for some exasperated CISO who fields so many dumb cybersecurity questions that he says FINE WE’RE STORING ALL OUR DATA ON THE MOON
For a large swathe of the population "OpenAI" and "AI" are the same, unfortunately.
January 24, 2025 at 1:49 PM
For a large swathe of the population "OpenAI" and "AI" are the same, unfortunately.
For us they are a useful local test proxy for our production cloud environment, which uses L4s. They still have their uses.
January 23, 2025 at 6:50 PM
For us they are a useful local test proxy for our production cloud environment, which uses L4s. They still have their uses.
Yep. This is not the build for tokens per second. Our use case for this is batch processing - it's not fast enough for real-time chat on larger models.
January 22, 2025 at 8:47 PM
Yep. This is not the build for tokens per second. Our use case for this is batch processing - it's not fast enough for real-time chat on larger models.
For us, it was buying a very old Dell PowerEdge server for £100 and putting 4 Tesla P40s in it.
Very slow, but it has 96GB VRAM and runs 70B models comfortably.
Very slow, but it has 96GB VRAM and runs 70B models comfortably.
January 22, 2025 at 8:35 PM
For us, it was buying a very old Dell PowerEdge server for £100 and putting 4 Tesla P40s in it.
Very slow, but it has 96GB VRAM and runs 70B models comfortably.
Very slow, but it has 96GB VRAM and runs 70B models comfortably.