junipertea

I believe specifically for Microsoft, they did bundle a default replacement for chatGPT in a lot of different places (Bing chat, Copilot) which use OpenAI models! But the end product is notably worse than native interface. There is a bare-minimum-level of usability required.

For chat apps, good enough is good enough. For something as universally useful and easy to use as ChatGPT, the bar is higher. I don't want to comment on the financial feasibility, but whatever Microsoft put out has been a complete flop even when free, making ChatGPT $8 subscription seem worth it in comparison

As a developer, I don't want apps to set up containers on my harddrive, but as a end-user it would be same

Honest answer: The pain of setting up an encrypted container. This seems simple, single use and immediately ready to use after downloading.

Maybe another company will take over at time. Why does one company have to stay perfect and on top of game for eternity?

The Silo is a book adaptation, not a new IP.

Can you elaborate on how you use it? I tried googling any mention of it and failed (all reddit posts were deleted). I installed it into sublime but it didn't seem to do anything

To write a blog post! "I used AI to solve this challenging problem" gets more traction than "I solved this challenging problem"

Most universities do not contain sufficient capacity to host both the conference and the overnight stays of thousands of people. Even a major university would face significant disruption by hosting a conference, it is simply not built for it. Building venues at each university doesn’t make sense either because it doesn’t have 100% utilization.

Conference venues provide incentives by having cheap venue rent and minor discount on housing and the city makes money by having more people do fun things in their down time.

Purely from efficiency, disregarding politics or “waste” or “delusion”, conference centers are built for a very rational reason - it makes sense and saves money for event organizers.

also, to answer the original question: yes they are

This bug can be mitigated via inflation.

The table row with 6k figure refers to high efficiency, not high compute mode. From the blog post:

Note: OpenAI has requested that we not publish the high-compute costs. The amount of compute was roughly 172x the low-compute configuration.

There is irony in “I don’t understand why people do X and ended up doing X but simpler”

Probably as copyright lawsuits settle and they will have to pay for their training data

There is Mamba 7B pre trained model on hugging face/ LM studio right now.

State size is constant (a matrix), attention is linear (a single unidirectional scan over all states)

Both statements are correct

I would assume all those companies want cutting edge (higher performance at lower energy cost), not chips that are 5-10 years behind. Those can barely run current android.

Doesn’t meta quest literally run on android?

Last time I tried anytype it automatically synced my notes somewhere with no option of opting out. The only option was to modify the source code and rebuild it.

Less data sovereignty than notepad.

Is the discussion about the paper, or about how it was unilaterally rejected?

How would you search through thousands of notes? Decrypt them all in memory, search and encrypt them again? How does this scale with attachments?

Would you download and host an awful book out of principle?

It corresponds to the paper formulas as a courtesy. While descriptive comments or function signature might be nicer, it still is reasonably easy to follow along from the paper description! (Although, most of the complexity is hidden away in selective_scan_cuda.fwd())

While neither of us can extrapolate behavior of “most people”, I and a lot of my middle class peers definitely this exact thought and acted on it.

unless it's way worse

What about PSU to power all of this, GPU motherboards, SSDs, CPU to not bottleneck this and sufficient RAM? You could argue it's more than 8, but definitely not 16. To utilize those GPUs to the max you need memory bandwidth to even feed it enough data.

Not LLM, just BERT, also did not actually outperform it.

source: https://kenschutte.com/gzip-knn-paper/

While there are some exception, majority of published deep learning research barely mentions statistics at all, it's optimization all the way down

To be fair, you can compile the code (including the protobuf format) and implement your own export to any format. It's a bit obtuse, but I had a obsidian vault full of markdown files that are also completely useless as they have a bunch metadata that needs the obsidian plugin to render. This is not a critique of obsidian, but there is a limit of markdown and obsidian is bypassing it using a lot of extra code. If obsidian got deleted from the internet, I would have my markdown files but a lot of the processing would be gone.

If we were to store markdowns only with linking and no advanced tables, do we need any advanced app on top?

1. Native performance (compared to Notion and Obsidian it's speedy) 2. UX (subjective)

While I would concur for any behavior (e.g. drinking), you don't really have a choice with being homosexual. Just don't be left-handed?