johntb86

7 ms·

31.
by johntb86 2y ago
I've found that OpenAI's Deep Research seems to be much better at this, including finding an obscure StackOverflow post that solved a problem I had, or finding travel wiki sites that actually answered questions I had around travel
32.
by johntb86 2y ago
I'd be curious what would happen if you SFTed a larger model with successful reasoning traces from the smaller model. Would it pick up the overall reasoning pattern, but be able to apply it to more cases?
33.
by johntb86 2y ago
Does anyone have an intuition about why looking at things in the frequency domain is helpful here? The DC term I can see, but I wouldn't expect the input data is periodic enough that other frequencies would be meaningful.
34.
by johntb86 2y ago
There are a lot of dimensions of appropriateness. For example it's inappropriate to respond to a question in English in Swahili, unless explicitly asked to. Or it's wrong to output complete gibberish. The model seems to have gener
35.
by johntb86 2y ago
Actual announcement: https://qwenlm.github.io/blog/qwen2.5-max/
36.
by johntb86 2y ago
Vulkan video isn't allowed to be enabled in Android Vulkan drivers, though perhaps that will change some day.
37.
by johntb86 2y ago
Who is cqwrteur? Is he a well-known character in the Linux community?
38.
by johntb86 2y ago
Yeah, a beacon isn't that useful unless you want to require that all pedestrians wear them as well. The only self-driving-car-related serious injuries/deaths I can remember are Uber hitting a pedestrian and Cruise running over a p
39.
by johntb86 2y ago
It seems like the ball bounces off the center of the paddle, not the edge, which always makes it look wrong. Maybe you're seeing the same problem?
40.
by johntb86 2y ago
Maybe people just don't realize how little pain you can feel with a needle. I switched from the old formulation of humira to the new one (smaller needle and no citrate) and the difference is night and day. Before, I was dreading it eve
41.
by johntb86 2y ago
https://www.quantamagazine.org/computer-scientists-prove-tha... gives the more layman-friendly version.
42.
by johntb86 2y ago
They're using TSMC 5-nm for WSE-3: https://spectrum.ieee.org/cerebras-chip-cs3
43.
by johntb86 2y ago
GPT-4o has seen a lot of examples of what people write about coding on the web, what code exists, and what tasks people want to do with that. But that general data doesn't include full process of coding - get a bug report, look at a co
44.
by johntb86 2y ago
LLMs can be funny. For example, look at Golden Gate Claude ( https://news.ycombinator.com/item?id=40459543 ). But they're not good at intentionally being funny, so we need to break them to get absurdist humor instead.
45.
by johntb86 2y ago
From looking at a longer video, it seems like a car honks when the car in front of it backs up.
46.
by johntb86 2y ago
Airbus airplanes have several different flight law modes (including direct law, which has no protections). So the biggest risk is probably that it switches laws and actually tries to carry out what you're doing.
47.
by johntb86 2y ago
Adding more transistors doesn't make your clock got faster, and it doesn't increase the speed of an individual transistor. The reason computers got both more transistors and faster in the past was that the transistors were continu
48.
by johntb86 2y ago
Maybe the extracted caffeine has some market value that can make up for some of the cost of extracting it.
49.
by johntb86 2y ago
By this point, instruction tuning should include tuning the model to use chain of thought in the appropriate circumstances.
50.
by johntb86 2y ago
>Sadly, I am unaware of Nest and/or Ecobee's support for this sort of a setup. Probably not cost effective for them. Ecobee has a setting to enable this: https://support.ecobee.com/s/articles/How-to-us
51.
by johntb86 2y ago
For these tokens you first need to unembed the result of the final layer, the re-embed the resulting token on the next pass. Has anyone investigated passing the raw output of one pass to the input of the next?
52.
by johntb86 2y ago
In theory the hardware should return corrupted video, return an error, or at least hang, but not anything worse. It's worse if the data structures specifying memory buffers are incorrect; then you may be able to read/write arbitra
53.
by johntb86 2y ago
They mention previous work on speculative decoding using similar techniques, but "ANPD dynamically generates draft outputs via an adaptive N-gram module using real-time statistics, after which the drafts are verified by the LLM. This c
54.
by johntb86 2y ago
What do you mean by saying that they're replaying signals from teleoperation demonstrations? Like in https://twitter.com/DannyDriess/status/1780270239185588732 , was someone demonstrating how to struggle to fo
55.
by johntb86 2y ago
Isn't all of recorded human history up until 2022 enough to train a pretty smart ai? It'll miss out on future trends, but some targeted training on trusted news sources should be enough. Eventually languages will change enough tha
56.
Nvidia's GPU IP Drives into MediaTek's Dimension Auto SoCs(anandtech.com)
1 points by johntb86 3y ago|0 comments
57.
Albania to speed up EU accession using ChatGPT(euractiv.com)
3 points by johntb86 3y ago|0 comments
58.
by johntb86 3y ago
Presumably with dram you also have to worry about refreshes, which can come along at arbitrary times relative to the workload.
59.
by johntb86 3y ago
I always liked the idea of slip lanes, as then you could check first for a conflict with a pedestrian (looking left and right), then cross that point and worry about a conflict with a car or bike (looking far left). But some people are now
60.
by johntb86 3y ago
Previous discussions: https://news.ycombinator.com/item?id=38912240 and https://news.ycombinator.com/item?id=38878780

More ›