lumost

47 ms·

931.
by lumost 3y ago
There is a weird assumption that Chinese tech corporations are permanently immune to American law. At most, this is an artifact of unidirectional free trade agreements written generations ago. Similar to Europe, I doubt US regulators would
932.
by lumost 3y ago
While the statement reads this way, and should be rewritten - the OPs underlying point is that as time goes by it becomes progressively less likely that a 6 month/2 year/5 year/10 year relationship has a distinct meaning for
933.
by lumost 3y ago
Speed is correlated with execution quality. All else being equal, A lab which is better equipped - with better access to materials, and motivated and competent researchers will deliver before the equivalent lab which is deficient in any o
934.
by lumost 3y ago
Maybe they are correlated :) GPT-4 "What types of materials have molecular structures which may lead to room temperature superconductivity?"
935.
by lumost 3y ago
I'm starting to think that we will see model fragmentation based on alignment preferences. There are clearly applications where alignment is necessary, and there appears to be use cases where people don't mind an occasionally fala
936.
by lumost 3y ago
I've recently wondered if we are under-estimating the burdens being placed on the working age population in most developed nations. While public benefits are one mechanism to support the aging population, we have also returned higher t
937.
by lumost 3y ago
Exploring a few options in my off-time. Main motivation for an OS LLM is to get it to do things which GPT-3.5/4 are somewhat promising at - but not good enough for applications.
938.
by lumost 3y ago
GPT-4 has a capped number of responses, it also costs $20/month. If it's only marginally better than GPT-3.5, why would I pay for it?
939.
by lumost 3y ago
anecdotally, it was better at generating code ~4 months ago. It makes one wonder if there is a market for "super-premium AIs" for software engineers/other technical users.
940.
by lumost 3y ago
Curiously, I observed that many investors (or at least people proposing to pitch investors) focus on the simplest of use cases. Perhaps the investor crowd is betting that they can consolidate a few of the successful stories? Or are simply l
941.
by lumost 3y ago
Regulation has vastly improved the situation for edibles. You can generally get accurate dosages with precise mixtures in any form factor you desire. However different states have diferring regulation.
942.
by lumost 3y ago
This will also assuredly fuel some interesting ideas about the origin of the Piri Reis map. A 120k year old ice sheet could mean that humans once lived on Antarctica. It does make one wonder if any ancient sea fearing humans happened to car
943.
by lumost 3y ago
California is mostly desert, national forest, and fire country. LA, and SD exist in their current state due to large scale water works that bring water from outside of California. For an American comparison, the northeastern US is home to 4
944.
by lumost 3y ago
I used to wonder why such impossible research projects could exist in a private business. To some extent, the existence of such a project can give credence that management is not oblivious to larger trends - and has a plan to leapfrog the c
945.
by lumost 3y ago
This a super interesting paper, but in my oppinion - they do not complete their core claim of offering a generalized mathematical structure for attention/recurrence. The specific structure they propose is very interesting and demonstra
946.
by lumost 3y ago
A good hypothesis to test, however why wouldn’t we observe human burials in the same (massive) cave system? It would seem odd to develop a ritualistic burial for pets and skip the burial for humans. Reaching the burial spot is extremely dif
947.
by lumost 3y ago
I mean, I was a GitHub copilot user - but I haven’t checked in such code. I use gpt-4, but I only commit my experimental code.
948.
by lumost 3y ago
I would be skeptical that "most" code is written by AI. Some code, certainly - but major enterprises are still terrified of the copyright outcome for existing non-local GPT implementations.
949.
by lumost 3y ago
Close! Was in a meeting and needed to see the behavior of a system when passed a string greater than 64 char. My mental capacity was used elsewhere - using chatGPT let me answer an important customer question authoritatively.
950.
by lumost 3y ago
Another way of thinking about it. If training on all the data yields a model which is functionally 5% better in online metrics, which would not be uncommon in a pareto distributed traffic pattern - then any subsequent partitioned model woul
951.
by lumost 3y ago
makes sense - I ultimately need to train the weights so was focusing on GPTQ, I'll try out ggml and see if the latency is better. I have some flexibility on whether I run inference/training on the same model instance. What context
952.
by lumost 3y ago
Most people grow skills over time. They are either going to be promoted and demonstrate their higher level skill - or work fewer hours. It’s hard to maintain a steady state organization, meta will need to find other incentives for individua
953.
by lumost 3y ago
Aye was on quantized weights using gptq.
954.
by lumost 3y ago
Close enough for the purpose at hand!
955.
by lumost 3y ago
I used GPT-4 to generate a non-cryptographic random 64 character string. It was faster to ask GPT-4 for the string than ask GPT-4 for the instructions to generate the string from my terminal. GPT-4 was faster than google.
956.
by lumost 3y ago
This wasn't so much overtraining, as the models learning something different than what we expected. If you look at a pixel by pixel representation of an image, textures tend to be more significant/unique patterns than shapes. Ther
957.
by lumost 3y ago
Thanks be to open-source https://huggingface.co/models?sort=trending&search=thebloke%... Has the quantized weights, available to download now. I tried out the Llama-2-7B-GPTQ on an A100 hosted at runpod.io. Llama-2 is a
958.
by lumost 3y ago
Having worked on ML products, there is sometimes debate on whether you should train on the test partition prior to prod deployment - after all, why would you ship a worse model to prod? Obviously you can't tell whether the model is
959.
by lumost 3y ago
This isn't an uncommon attitude, tech has just had the unique ability to re:invent itself and build defensive moats faster than competition can catch up.
960.
by lumost 3y ago
aye - my assertion is that the limits of the human body are not a major limiter in current military aviation. You can see an example of how current military planners are thinking about things with the B-1R concept (ultimately rejected, due

More ›