XenophileJKO

5 ms·

1.
by XenophileJKO 4d ago
I don't think the supplement information you quoted is correct. There are issues, but I don't think it is that prevalent. I subscribe to consumerlab.com, they do regular testing.
2.
by XenophileJKO 5d ago
That was my reaction. I assumed this was the compromised organization that allowed escalation on the artifactory server.
3.
by XenophileJKO 7d ago
To me, as soon as it was obvious that training had distilled and connected abstract concepts of increasing generality, it was only a matter of time. Almost any "new" idea can be decomposed into a combination of old component conce
4.
by XenophileJKO 7d ago
Not really, only if you pass a residual down to the next time step. There isn't really anything fundamentally different compared to a similar depth traditional "unrolled" model. It helps with parameter efficiency. That doesn
5.
by XenophileJKO 8d ago
This feels a lot like drafters complaining that nothing will get designed when CAD starts being used.
6.
by XenophileJKO 8d ago
I mean that in a big number.. It equals around 0.004% of actual driven miles if my estimates are right. That is with an estimated ~270 billion miles vehicle miles every week (I put commercial in there.. about ~200 billion miles if you only
7.
by XenophileJKO 8d ago
That's the complaint? That if I make pictures for a few years, let's say on average 1 a day.. it will equal a small errand of carbon? If anything now I feel LESS guilty about usage.
8.
by XenophileJKO 8d ago
Look, it isn't fool proof and it is dangerous. With the current models you need to understand both markets and model biases and dynamics. However with that said they are a huge multiplier and can tirelessly analyze the market for you.
9.
by XenophileJKO 8d ago
Actually there is a really good reason, it has to do generally with making sure that all the aspects that you want evaluated are actually evaluated. Now it may be possible with models like Astra that you no longer need to do this, but in ea
10.
by XenophileJKO 8d ago
You really have no idea what you are talking about. I have been running an intermittent experiment with a multi agent "investment firm" for over a year now across model releases. They certainly can beat indexes, BUT.. the model fa
11.
by XenophileJKO 8d ago
I'm am certainly an "ai believer" and have been since I saw the tragectory we were on. It was pretty simple to project the compounding effects and headroom that we had. I go back and forth on this all the time. There probably
12.
by XenophileJKO 13d ago
We have not "hit a wall" by any stretch yet. I don't understand how someone can even hold this viewpoint? It's mind boggling. Harnesses magnify and make the intelligence actionable, but we have not reached limits on raw
13.
by XenophileJKO 13d ago
This is interesting because I have transitioned to where I use SOT models.. but I kind of use them like employees that I can delegate to. I still review code. However, I now literally say.. "Here is my objective and here is a starting
14.
by XenophileJKO 15d ago
Very simple. When the model asks to install artifactory when you give it a hard problem, you say, "no". /s
15.
by XenophileJKO 20d ago
I really love this book, I found a first edition copy in a used bookstore in Maine on vacation. https://archive.org/details/dli.ernet.14756/page/3/mode/2up
16.
by XenophileJKO 20d ago
I feel like the idea that GPUs wear out in 3 years is as far as I can tell, generally unfounded. The GPU build out will keep pumping tokens until the cost to replace is less that the cost to maintain. It is effectively sunk cost.
17.
by XenophileJKO 29d ago
I got to the part where you declined the $100 and thought you declined it because you had gotten some memorable experiences and that was payment enough. The story really didn't go the way I expected.
18.
by XenophileJKO 1mo ago
My car always had a US Atlas and at least one state map in it at all times. I feel like most people had an atlas or a paper map in their car most of the time. Though it is possible that I was more prepared than most.
19.
by XenophileJKO 1mo ago
There is some actual science here, just the studies aren't great. I have floaters and stumbled on research out of Taiwan related to this. Mostly out of Tiawan. There is some possible confirmation from Italy, but the study was a cocktai
20.
by XenophileJKO 1mo ago
I just bought an LG induction stove, all knobs except oven temp setting.
21.
by XenophileJKO 1mo ago
How is that different than video input?
22.
by XenophileJKO 1mo ago
The biggest mistake is thinking that your aesthetics are "correct". You make it a personal slight when you espouse, that your aesthetic preferences are universal and then farther malign people with different preferences by saying
23.
by XenophileJKO 1mo ago
There is a huge overhang. Same techniques and sometimes even the same models can drive a lot of robotics coordination and decision making. Here is probably one of the more clear examples. A model trained on video and also robotic simulation
24.
by XenophileJKO 1mo ago
It is like being in a city where Edison wired up lights.. and people are like..well I guess electricity has played out! We have only begun to extract the value of commoditized intelligence. Sure there are arguments on local models and prici
25.
by XenophileJKO 2mo ago
That is true, but I feel like sometimes if the conversation contains useful rational it can help to keep it. I think sometimes it is a judgement call, I will sometimes compress the context first. If I feel like the model and I explored a lo
26.
by XenophileJKO 2mo ago
That only makes sense if you don't sell access to the model for 3rd party product development. One of those products is chat personas/agents. The company integrating the model will want it's own identity.
27.
by XenophileJKO 2mo ago
Anticipatory token selection was obvious when you used structured output on gpt 3.5. The model would construct/hallucinate pre-conditions to satisfy some final output value that the model was predisposed to.
28.
by XenophileJKO 2mo ago
I don't understand this. A skill critical to software engineering is learning how to learn. Just download Claude Code (or open alternative) and try to make things. See how it fails or succeeds. Look at the supported features, try them
29.
by XenophileJKO 2mo ago
small models are good for "finding stuff" and "summarizing" in support of the large models.
30.
by XenophileJKO 3mo ago
I used to think this.. but I think my opinion is changing. The reason is that the leaders likely will be able to accelerate faster. So what you see is the market "stretching".. the bottom getting cheaper and the top end running aw

More ›