panabee

8 ms·

31.
by panabee 2y ago
It's unclear why this drew downvotes, but to reiterate, the comment merely highlights historical facts about the CUDA moat and deliberately refrains from assertions about NVDA's long-term prospects or that the CUDA moat is unbreac
32.
by panabee 2y ago
Alternatives exist, especially for mature and simple models. The point isn't that Nvidia has 100% market share, but rather that they command the most lucrative segment and none of these big spenders have found a way to quit their Nvidi
33.
by panabee 2y ago
Google was omitted because they own the hardware and the models, but in retrospect, they represent a proof point nearly as compelling as OpenAI. Thanks for the comment. Google has leading models operating on leading hardware, backed by soph
34.
by panabee 2y ago
Nvidia (NVDA) generates revenue with hardware, but digs moats with software. The CUDA moat is widely unappreciated and misunderstood. Dethroning Nvidia demands more than SOTA hardware. OpenAI, Meta, Google, AWS, AMD, and others have long fa
35.
by panabee 2y ago
To address the downvotes, this comment isn't guaranteeing OAI's success. It merely notes the remarkably elevated probability of OAI escaping Nadella's grip, which was nearly unfathomable 12 months ago. Even after breaking fre
36.
by panabee 2y ago
Nadella is a superb CEO, inarguably among the best of his generation. He believed in OpenAI when no one else did and deserves acclaim for this brilliant investment. But his "below them, above them, around them" quote on OpenAI may
37.
by panabee 2y ago
This is a good question and encapsulates the challenges of food and drug regulation. Yes and no. At certain concentrations, many safe compounds become dangerous in humans. Even at tiny doses, foods like peanuts may be safe for the vast majo
38.
by panabee 2y ago
Balancing safety and innovation in human health is incredibly difficult. This isn't a criticism of the FDA, but rather an observation of facts underscoring both the challenges and opportunities. Some foods consumed safely by humans are
39.
by panabee 2y ago
With many research areas converging to comparable levels, the most critical piece is arguably vertical integration and forgoing the Nvidia tax. They haven't wielded this advantage as powerfully as possible, but changes here could signa
40.
by panabee 2y ago
Amazon is a retailer and strives to offer choice, whether of books or compute services. AWS is the golden goose. If Amazon doesn't tie up Anthropic, AWS customers who need a SOTA LLM will spend on Azure or GCP. Think of Anthropic as th
41.
by panabee 2y ago
This is unrelated to your main point, more to clarify an edge case for those interested in learning more: B cells can breach the BBB under certain conditions. At least one pathogen, Epstein-Barr virus, is known to inhabit B cells.
42.
by panabee 2y ago
Proteins can cause DNA alterations via chromosomal translocations. If these occur in germ cells, the new sequences may transmit permanently to future generations. Depending on the definition of "information transfer" and the willi
43.
by panabee 2y ago
a doctor friend highlighted two key limitations: only six cases were evaluated per physician and half the physicians were only residents.
44.
by panabee 2y ago
thanks for sharing. the implications are fascinating, if the findings are generalizable and reproducible. the study suggests LLMs may already be materially superior to experts in a critical field like medicine, and that inexpert users hold
45.
DiLoCo: Distributed Low-Communication Training of Language Models(huggingface.co)
3 points by panabee 2y ago|0 comments
46.
Revised Chinchilla scaling laws – LLM compute and token requirements(educatingsilicon.com)
1 points by panabee 2y ago|0 comments
47.
Compute-Optimal Context Size(manifestai.com)
3 points by panabee 2y ago|0 comments
48.
An Empirical Study of Mamba-Based Language Models(arxiv.org)
43 points by panabee 2y ago|3 comments
49.
Human Language Understanding and Reasoning (2022)(amacad.org)
1 points by panabee 2y ago|0 comments
50.
by panabee 2y ago
First paragraph: Mapreduce is a programming model for processing and generating large data sets.4 Users specify a map function that processes a key/value pair to generate a set of intermediate key/value pairs and a reduce function
51.
MapReduce: A Flexible Data Processing Tool (2010)(cacm.acm.org)
3 points by panabee 2y ago|1 comments
52.
RLHF: Reinforcement Learning from Human Feedback(huyenchip.com)
1 points by panabee 2y ago|0 comments
53.
Gemini 1.5 Model Family: Technical Report [pdf](storage.googleapis.com)
57 points by panabee 2y ago|3 comments
54.
by panabee 2y ago
Audio generation is currently in private beta. There are only a few papers publicly available, but hopefully Google expands the list soon.
55.
Illuminate: Turn academic papers into AI-generated audio discussions(illuminate.withgoogle.com)
3 points by panabee 2y ago|1 comments
56.
Sparse Llama: 70% Smaller, 3x Faster, Full Accuracy(cerebras.net)
40 points by panabee 2y ago|1 comments
57.
by panabee 2y ago
interesting observation and experience. must have made thesis development complex, assuming the realization dawned on you during the phd. what do you trust more than NMR? AF's dependence on MSAs also seems sub-optimal; curious to hear
58.
by panabee 2y ago
this is very astute, not only about deepmind but about science and humanity overall. what CASP did was narrowly scope a hard problem, provided clear rules and metrics for evaluating participants, and offered a regular forum in which candida
59.
Mitochondria and Chloroplasts(khanacademy.org)
2 points by panabee 2y ago|0 comments
60.
by panabee 2y ago
this is correct. it's clear that smoking elevates cancer risk, but why doesn't it cause cancer in all smokers? to develop a cure, we must better understand the causal mechanisms. this starts with acknowledging what we know and don

More ›