gdiamos

13 ms·

181.
by gdiamos 2y ago
It's popular enough that it causes pain for a lot of people. Coming from C++, IMO, it is vastly better.
182.
by gdiamos 2y ago
This is really easy to set up - and is much more accurate than asking the LLM to predict True/False. Just feed the outputs of an embedding API into logistic regression, e.g. from sklearn. import voyageai vo = voyageai.Client()
183.
by gdiamos 2y ago
Nice to see this finally after 15 years.
184.
by gdiamos 2y ago
This pattern is used in many places. Build systems. Now compiler passes. Container layers. Also in ML systems, eg dvc. Where else is it missing?
185.
by gdiamos 2y ago
Memory locality depends on your perspective. The CPU would always be slower if the data originated in GPU memory.
186.
by gdiamos 2y ago
How is it going to get enough money to pay for the computers to run the LLMs? Steal it?
187.
by gdiamos 2y ago
There are clear flaws. I haven’t figured it out. Being able to form and disband teams dynamically is an advantage. That does seem to favor a big enough pool of people in the same place. Or relying on remote for projects spanning pods.
188.
by gdiamos 2y ago
My favorite environment is a 5-10 person office, no more than 5 miles from everyone’s home. I know how to make that work in a startup, but its premise doesn’t scale beyond a pizza box. Has anyone figured out a better way that works at 100 p
189.
by gdiamos 2y ago
Nice, I wonder if there is a way to make it conditional, e.g. to pick a different key for a cpu vs cuda build.
190.
by gdiamos 2y ago
uv significantly speeds up my pytorch in docker builds # Setup virtual env ENV VIRTUAL_ENV=/app/.venv ENV PATH="$VIRTUAL_ENV/bin:$PATH" RUN python3 -m venv $VIRTUAL_ENV RUN . $VIRTUAL_ENV/bin/a
191.
by gdiamos 2y ago
I'm so curious to see some multi-agent systems running with inference this fast.
192.
by gdiamos 2y ago
just in time for their ipo
193.
by gdiamos 2y ago
It’s approximately bad, like most of ML On one side: Would you expect a model trained on no Spanish data to do well on Spanish? On the other: Is it okay to train on the MMLU test set?
194.
by gdiamos 2y ago
I personally hire people I’ve worked with before and I know what they can do. If I can’t do that, I ask to see their prior work. Good programmers have usually written a lot of code. Having them walk through and explain it usually gives a go
195.
by gdiamos 2y ago
This is exactly right. “ If you want them to MAYBE change, talk to them directly. If you want them to get stabbed by management, put it in their performance review. Anything you say can and will be used against... your co-workers.” Doing th
196.
by gdiamos 2y ago
Github was an early OpenAI design partner. OpenAI developed a custom LLM for them. It's so interesting that even after that early mover advantage they have to go back to the foundation model providers. Does this mean that future tech
197.
by gdiamos 2y ago
Hire and retain only the best
198.
by gdiamos 2y ago
Glad to see Sean Baxter is working on this
199.
by gdiamos 2y ago
I am a founder. It’s not meant to be a secret. Obviously I’m biased, but I also spend every day using tools like this. Regarding json compliance, we have a formal grammar and a test suite. If you find a bug please report it. I’d appreciate
200.
by gdiamos 2y ago
I usually come to a different conclusion using the JSON output on Lamini, e.g. even with Llama 3.2 3B https://lamini-ai.github.io/inference/json_output Most of these models can read. If the relevant facts are in the pr
201.
by gdiamos 2y ago
Still waiting for a competitive diffusion llm
202.
by gdiamos 2y ago
After deep diving into building datacenters, I realized that traditional clouds (AWS, Google, azure) add a huge margin to GPUs like H100s. You can (and I think should), cut it out by going with coreweave, lambda, another or doing it yoursel
203.
by gdiamos 2y ago
Can it modify its training data?
204.
by gdiamos 2y ago
H100s are available at increasingly affordable prices
205.
by gdiamos 2y ago
How much of this is specific to control loops that execute at 60hz?
206.
by gdiamos 2y ago
Nice to see OpenAI adopting useful Claude features
207.
by gdiamos 2y ago
RNNs always had better scaling law curves than transformers. BPTT was their problem
208.
by gdiamos 2y ago
Cerebras has a real technical advantage in development of wafer scale.
209.
by gdiamos 2y ago
why not tell us? Or point out what is actually false? You rebutted the points that the flop limit will not actually limit anyone by saying that gpt4 is out of reach of startups. Is OpenAI a startup? Is Anthropic? Is Grok? Is Perplexity?
210.
by gdiamos 2y ago
Claude, SSI, Grok, GPT, Llama, … Should we crown one the king? Or perhaps it is better to let them compete? Perhaps advanced AI capability will motivate advanced AI safety capability?

More ›