msoad

8 ms·

91.
by msoad 2y ago
> The technology isn’t there, but they have a deadline The technology is there, as demonstrated by OpenAI's ChatGPT Voice Mode. With the resources and talent that Amazon possesses, they should at least be able to demo something simi
92.
by msoad 2y ago
MMLU PRO is the benchmark I trust the most. I noticed they are using 5 shots and CoT. Is that true for GPT4 and Sonnet as well?
93.
by msoad 2y ago
Kinda ironic that the article itself was updated
94.
by msoad 2y ago
As this article argues, problem with CSS Classes is that a CSS Class can be applied to _any_ element where custom properties will be scoped to the elements that the CSS was meant to be used for.
95.
by msoad 2y ago
Sparse checkout is a new Git feature that can help with this
96.
by msoad 2y ago
This is very impressive! This is a risky bet for actual production use. There is no way you can 100% be sure that the generated binary is correct and matching behavior with the original binary. There are cases that don't have unit test
97.
by msoad 2y ago
Phi-3 blow this out of the water. Benchmark | Gemma 2 (9B) | Phi-3 Small (7B) -----------------------------|----------------|------------------- MMLU (5-Shot) | 63.6 | 75.7
98.
by msoad 2y ago
Americans do not comprehend how advanced BYD and others are in EV. It's no accident that their government is fighting the import of Chinese cars
99.
by msoad 2y ago
You can play around with the language itself here https://polarity-lang.github.io/oopsla24/#ChurchEncodingCoda...
100.
by msoad 2y ago
https://colab.research.google.com/drive/1VYm3hOcvCpbGiqKZb14...
101.
by msoad 2y ago
It's just on brand to hate on landlords. You don't need logic for that.
102.
by msoad 2y ago
This article from Berlin is very interesting. The city presents two totally different universes when it comes to renting an apartment. Either you're one of the lucky ones to secure a rent-controlled unit and never (literally never) off
103.
by msoad 2y ago
That makes sense!
104.
by msoad 2y ago
The article mentions that Apple can use OpenAI responses for further training their own models. I don’t want to say that’s impossible but that means they really screwed OpenAI in this deal. OpenAI terms don’t allow training on responses
105.
by msoad 2y ago
Almost 1% flowers!
106.
by msoad 2y ago
Are we plateauing with those LLM benchmarks?
107.
by msoad 2y ago
Microsoft just announced that they are getting better price/performance ratios with AMD for inference. It's happening already. Most of GPU thirst for training at the moment but eventually most of workload will be inference. And th
108.
by msoad 2y ago
> how do you argue that these models are not able to reason? I don't make this argument. Benchmarks like CLUTRR[1] show how poorly LLMs do in reasoning. [1] https://github.com/facebookresearch/clutrr
109.
by msoad 2y ago
It seems like a hack to be honest. Problem at hand is not to make transformers do addition of 100 digit numbers. Problem is the current systems can’t reason about things, math included. Optimizing for a certain use case is not gonna take us
110.
by msoad 2y ago
It's as if the leaders who made this decision to put LLM generated content in front page of Google search results never used LLMs. Put this into any LLM and see how they go on hallucinating hard[1]: "in the book "wisdom of Ma
111.
by msoad 2y ago
I don’t want to be a fanboy but the wallpaper is so soulless and industrial. I like how Apple always tries to bring nature to computing with wallpapers and screensavers. For example https://iso.500px.com/iphone-6-milky-way-w
112.
by msoad 2y ago
homelessness has been on the rise
113.
by msoad 2y ago
With all of those stories coming out of OpenAI I'm happy I didn't join them. A lot of sleazy and shady practices.
114.
by msoad 2y ago
Compared to Mirasol3B[1] this is not supporting audio as a modality. What Google has done with Mirasol3B made the demo of "Astro" in Google I/O possible. They do a little of cheating by converting audio to images(spectrogram)
115.
by msoad 2y ago
Berlin famously casted a gender neutral voice for public transport announcements. It would be nice if OpenAI would have done that too. At least for one of the options. Computers better not have genders imo
116.
by msoad 2y ago
Before next token prediction become the only meaning of “generation” generative art was a way of describing art automatically being generated. It never sounded good so we always thought of it as a visual medium.
117.
by msoad 2y ago
xLSTM is really really hard to grasp. It's mind boggling how they came up with such a unique architecture. I read the paper multiple times and still not sure if I can say I understood it.
118.
by msoad 2y ago
It’s not just OpenAI. I think AI winter is coming much sooner than expected. Throwing more compute at the current solution is not yielding better results it seems. Reasoning will not be solved by a bigger LLM
119.
by msoad 2y ago
It took me so long to fully appreciate TypeScript's design decision for doing structural typing vs. nominal typing. In all scenarios, including the "issue" highlighted in this article there is no reason for wanting nominal ty
120.
by msoad 2y ago
They are admitting[1] that the new model is the gpt2-chatbot that we have seen before[2]. As many highlighted there, the model is not an improvement like GPT3->GPT4. I tested a bunch of programming stuff and it was not that much better.

More ›