lossolo

53 ms·

901.
by lossolo 4y ago
All of those are inputs and you can restrict them the same way you restrict a text dataset for training to achieve the results you desire - a model that leans right or left, one that wins a game using a certain strategy, or generates code u
902.
by lossolo 4y ago
This submission should be flagged, as there is no evidence to support the claim in the article. The information provided only shows that a system process called "medianalysisd" attempts to open a TCP connection to "api.smooth
903.
by lossolo 4y ago
I had additional context in that chat session and whole/part of the whole session is used as input for every output. I think it was fine-tuned on that example from the paper, that's why it can answer it in isolation. This is inter
904.
by lossolo 4y ago
Does it lead to anything good if you will not stop it at certain point? Will you not end up with pure nihilism if it does not end?
905.
by lossolo 4y ago
And answer from the new version: > After using 10 chairs, we have 19 chairs - 10 chairs = 9 chairs in the room. Then after buying 9 more chairs, we have 9 chairs + 9 chairs = 18 chairs in the room.
906.
by lossolo 4y ago
That's a bad example, it's fully rational and chatGPT gets that "It's still the same beer but the location and atmosphere of the resort may justify the higher price.", you are not paying only for beer there, you hav
907.
by lossolo 4y ago
> Additionally new research is uncovering serious and significant risks for the vaccine with fairly marginal protection benefits. Could you provide a link to that research ?
908.
by lossolo 4y ago
A large amount of data is not an issue, but obtaining a high amount of high-quality data is challenging. This is why open-source models do not perform as well as GPT-3 models in real-world usage.
909.
by lossolo 4y ago
> If you specifically mean a general LLM trained on a general language corpus with instruction finetuning this is correct. Yes, thanks, that's what I meant. > If you are training a LLM on a domain specific corpus or finetuning on
910.
by lossolo 4y ago
It's not just about compute; if that were the case, then models like BLOOM and OPT, which also have 175 billion parameters, would have the same performance for real-world use cases as GPT-3, but they don't. Datasets are also very
911.
by lossolo 4y ago
Depends on precision, you can run ~5B model with fp32 precision or ~11B fp16 model max. Int8 is really bad for real world use case so not mentioning it. But if you are looking to get performance of ChatGPT or GPT-3 then don't waste you
912.
by lossolo 4y ago
Indeed, it is like comparing the speech of a 2-year-old child to that of a college professor.
913.
by lossolo 4y ago
> reproduces GPT-2 (124M) on OpenWebText, running on a single 8XA100 40GB node in 38 hours of training For comparison GPT-3 has more than 1000x more params (175B) and training time was around 2 months on ~1500 V100 GPUs which costs milli
914.
by lossolo 4y ago
I think OP meant SD model, it's 4 GB. ChatGPT 175B version is ~350+ GB.
915.
by lossolo 4y ago
Thanks, just skimmed the paper, I think that "they automated RLHF" statement is maybe too strong here, there is still manual process but it seems like they optimized away a lot of manual labeling work.
916.
by lossolo 4y ago
I think the problem is time. To replicate the OpenAI RLHF architecture, they need their own high-quality dataset, which takes time to create. Without details about the hyperparameters, RL architecture, and omitted steps, they need to test a
917.
by lossolo 4y ago
It can be replicated, it's basically clever fine-tuning of GPT-3 with human assistance. Google has model that can do what ChatGPT can for some time already.
918.
by lossolo 4y ago
It's not about 100 easy autopilot trips, it's about 1001 trip where you as human can easily see what to do but AI will interpret reality wrong because it was not trained to handle that edge case and you will die.
919.
by lossolo 4y ago
In my case, all of TLS connections are new and from real clients (not a benchmark/test)
920.
by lossolo 4y ago
All dynamic content, all hitting data storage. There are no simulated clients, this is all real traffic from real clients, a lot of requests do writes, some do only reads.
921.
by lossolo 4y ago
For offloading SSL I use haproxy with some custom settings and a few non standard kernel settings.
922.
by lossolo 4y ago
TLS handling would dominate your performance, kernel bypassing would not help here unless you would also do TLS NIC offloading, you still need to process new TLS sessions from OP example and they would dominate your http processing time (ex
923.
by lossolo 4y ago
> Use HTTPS and watch it fall down to only 400 req / sec under load test [ without connection: keep-alive ]. I'm running about 2000 requests/s in one of my real-world production systems. All of the requests are without kee
924.
by lossolo 4y ago
It seems like an article from Kodak or Nokia. It's only a matter of time, and it will be sooner rather than later, that multimodal LLMs will replace search engines. All of the problems mentioned in the article are being addressed as we
925.
by lossolo 4y ago
COT is so yesterday! SOTA is LAMBADA[1] aka backward chaining also from Google, that significantly outperforms chain of thought and select inference in terms of prediction accuracy and proof accuracy. [1] https://arxiv.org/a
926.
by lossolo 4y ago
It doesn't always help, it's about how is epoch defined in code, even on 64 bit system it can be defined as int32 and will overflow.
927.
by lossolo 4y ago
Max int32 is 2147483647 and minimum int32 is -2147483647, if you overflow you go to the beginning of the range so to -2147483647, if you subtract 2147483647 seconds from January 1 1970 you will arrive at 13 December 1901.
928.
by lossolo 4y ago
It's not housing costs for most societies in developed countries, it's keeping up with joneses , which constantly change the meaning of "standard way of living". 1000 years ago most people would kill to live in standard
929.
by lossolo 4y ago
It's about keeping up with joneses more than an specific item like iPhone, which is just an example of this phenomenon.
930.
by lossolo 4y ago
> Anything that has your attention can include ads. It's funny because paper[1] from Google that started transformers and LLMs is called "Attention is all you need". And I agree that you can easily incorporate ads into cha

More ›