benob

13 ms·

151.
by benob 3y ago
I am trying to find support for your last argument. https://paperswithcode.com/sota/common-sense-reasoning-on-ar... GPT-3 53.2 GPT-3.5 85.2 LLaMa-65B 56.0 Any idea of the performance of an instru
152.
by benob 3y ago
Probably. Check the kosmos-1 paper from Microsoft that appeared a few days before GPT4 was released: https://arxiv.org/abs/2302.14045
153.
by benob 3y ago
> Me: What are some questions you cannot answer? > FastGPT: I apologize, but I do not actually have the ability to determine what questions I cannot answer. I am Claude, an AI assistant created by Anthropic. Good to know ;)
154.
by benob 3y ago
Why isn't there a reddit clone ready to host all those communities under a different economic model (such as donations)?
155.
by benob 3y ago
Anything similar for piano?
156.
by benob 3y ago
[1] is a good start, although if you want to train from scratch on CPU, you'll have to downscale as transformers need quite a bit of data before they learn to use position embeddings. For example try a single-layer RNN on Shakespeare t
157.
by benob 3y ago
The model tries to predict what the next word will be given a context, typically a question to be answered or a text to be continued. It does so by looking at how each pair of words from the context contribute to this prediction. That'
158.
by benob 3y ago
An ironic version could display the weather forecast.
159.
by benob 3y ago
How about going one step further and constrain transformer output with a context-free grammar? That way you can generate more conformant code such as Python or C.
160.
by benob 4y ago
Yes you can finetune the model with reference output for any kind of language task. For translation, you are better off starting from a model specifically trained for that purpose such as facebook/nllb-200-distilled-1.3B. It will be fa
161.
by benob 4y ago
They probably use something similar to Kosmos-1 ( https://arxiv.org/abs/2302.14045 ): Encode images as vectors with something like CLIP, then map them to the token space and input them between <image> </ima
162.
by benob 4y ago
Picked the wrong one. LoRA, Low-rank Adaptation of LLMs ( https://arxiv.org/pdf/2106.09685.pdf ), consists in adapting the weights of a big neural network to a target task (here, answering to instructions). It doesn&#x27
163.
by benob 4y ago
It's probably that by optimizing code they reduced CPU load, therefore reducing fan noise due to overheating, and finally inducing clearer sound quality with lower SNR. It makes sense to me.
164.
by benob 4y ago
After using zig for a few projects, it's amazing how it feels similar to C when coding.
165.
by benob 4y ago
I wish there was a cheaper option without the AI feature (that is probably what costs them the most).
166.
by benob 4y ago
How long before subtle prompts that generate "sudo rm -rf /" as a command are proposed?
167.
by benob 4y ago
I hate those websites that mess-up with browser history...
168.
by benob 4y ago
Here is what I came up with for calling a c function after poking around a little bit: #include "pocketpy.h" // builtin function that can be called PyVar test(VM* vm, const pkpy::Args& args) { print
169.
by benob 4y ago
How does this compare to tinypy? [1] http://www.tinypy.org/
170.
by benob 4y ago
People have been grounding LLMs with internet searches for a few months: WebGPT: Browser-assisted question-answering with human feedback https://arxiv.org/abs/2112.09332
171.
by benob 4y ago
Maybe culture has shifted to other areas than music, movies, performing arts, etc as those have become entertainment business opportunities.
172.
by benob 4y ago
From the abstract: "This paves the way for low-cost, broadly accessible, and privacy-preserving algorithms for human sensing." How much do we have to wait for the follow up paper: "Human identification through wifi signals&qu
173.
by benob 4y ago
Can we interpret that as good news for opensource drivers?
174.
by benob 4y ago
Could the os restrict a range of ports to a non-root user? That would cover your concerns, wouldn't it?
175.
by benob 4y ago
What is the license for those pictures? Neither lexica nor openart talk about it.
176.
by benob 4y ago
According to the wikipedia page of the Playdate, it's run by a STM32F746 microcontroller [1]. It is interesting that it has lower specs than the Mario Game and Watch STM32H7B0 cpu [2], which is itself quite limited by its RAM and flash
177.
by benob 4y ago
Is there a way to embed redundancy in a crypto key so that another key a few bits away can still decrypt the data?
178.
by benob 4y ago
orjson, the fast underlying json parser, is written in rust.
179.
by benob 5y ago
I'd love to try that 127 bpp display they mention.
180.
by benob 5y ago
For some time I though it was all about the security implications of strtok().

More ›