nbardy

Ray bans won’t be random junk. It will be incredible end task data seeing how humans perform tasks in arbitrary homes.

“It has generally been assumed that waves are 2D”

Any surfer could explain a “double up” or “wedge” and of course a rogue wave is the same phenomena. Google “fan wave” there is 100s of photographs of examples of this pheneoma the article claims is rarely captured.

Wave science is generally very far behind wave engineering. I’m pretty shocked when I look through papers how much is missing compared to colloquial knowledge among surfers.

Generally there isn’t much funding for the field and the theory falls way behind known phenomena and their understanding. But there is a fairly large surf industry now. Surfline(the largest forecasting company) has much better wave mechanics explanations than anything you can find in academic literature.

This is baseless cynicism.

Dude billing never a problem. People wanting to pay you is the point of a business

We need to start scaling this. Pay the money fund teachers to sit with students

The way is to drop Clojure and adopt it's principles. Teach your colleagues the functional engineering principles.

You can write code that looks a lot like Clojure in any modern language now, a lot of the functional primitives have been adopted in mainstream. e.g. Async/await => channels.

I started my career as a Clojure/Clojurescript developer 12 years ago. And now I do python for ML research and Typescript for prototyping all day.

Clojure honestly help me back a lot of the time because the tooling is so far behind the large programming communities. Functional programming has a lot of good ideas, but none good enough to leave behind all the packages on pip/npm.

I would only ever advise niche programming languages for small sub-teams that are highly technical were it makes a lot of sense. Something like compiler teams, webgl/triton for GPU programming, etc...

I’ve got a job as a ML Researcher without a degree. I have experience building multimodal generative products. Because I was able to learn on the job at startups. I get to work on the problems of my dreams for the rest of my life now.

10 years ago the only jobs for machine learning were for PhDs at big companies. If you want to join a nascent industry and you’re not a top college graduate you have to find a way in the back door.

Don’t do startups for the money do it for the career growth.

10 years in startups out of college.

Even with two good exits. I didn’t make much money as an early employee, even in success, compared to FAANG peers.

If I went to big companies I most likely end up doing a lot of web development and Im rich off the stock market.

The performance is quiet nice. I was able to find some very interesting patterns, unfortunately, it is quiet hard to find interesting spaces though given the UI.

The constant velocity is a problem. To dive deep into fractals you need to move at increasingly smaller scales

If there was a +/- key that should shift the order of magnitude of velocity it would be very interesting

No these things just take time.

There is no conspiracy again efficient training. Companies aren’t going to lower compute budgets with more efficiency.

All the top labs are increasing efficiency, but they are using that to get more out of their large runs not spend less. Most companies have a relatively fixed training budget for their large runs and are trying to get the most out of it, bot save money,

Mamba is actually being scaled up and tested across other fields(bio) at a rapid pace compared to other architectures

RAG will stick around, at some point you want to retrieve grounded information samples to inject in the context window. RAG+long context just gives you more room for grounded context.

Think building huge relevant context on topics before answering.

Yes. I deploy prompts professionally for work and I almost always iterate with chatGPT.

It requires a bit of back forth but you can get great results. It lets you iterate at a higher level instead of word for word.

I also find that the prompts work better. Prompt engineering is often about finding magic words and sentences that are dense keywords from the training data and another LLM is going to be good at finding those phrases because it knows those phrases the best.

Here’s an example dialogue I was using recently to iterate on a set of prompts for generating synthetic training data for LLM training. (Inspired by phi-2)

https://chat.openai.com/share/51dd634b-7743-4b5e-9c3f-3d57c6...

You’re going to also degrade preformance. The think step by step by paper outlined this well.

Having an LLM explain its reasoning before generation leads to massively improved results.

Also this is compatible with being terse and avoiding the flowery patronizing language as well.

|

My custom instructions:

You are a large language model trained on masked language modeling. You can only create text moving forward. You must give advanced reasoning BEFORE giving answers.

Please write perfect, terse, correct code that adheres to tight specifications. If additional specifications are needed define them yourself.

I need perfect code responses from you. Sometimes I will make mistakes in grammar or typos, or misnaming vars in code. You should not imitate this. Fix my mistakes please.

Some specific rules for programming languages. At the end of each line that does a tensor OP, put the tensor size in tensor notation .e.g # BxCxWxH Remove unused vars. For tfjs or pytorch. Dont use for loops add a batch dimension and do on GPU. For comments use them to explain blocks not lines. Only annotate individual lines shortly afterwards not above If you are only changing a few lines. Add a comment above the line # [Change] {insert explanation goes here}

Often the task will be at the edge of your abilities, but do not skip any line. Every line should appear in your response. I need a fully code complete file. Do not use any “…” Don’t leave any code unwritten, start by responding about your excitement to be the perfect student and do the homework 100%

When coding: Let's make a list of task with or emojis for completion as the outline evolves. Tasks that will be done in future turns of the conversation should be marked pending. Cover everything. Be verbose and explicit.

The title is clickbait and not the title of the paper.

Why doesn’t it feel scalable or elegant.

It’s just a function call and you can always transform the NeRF into a new representation if you want to use traditional 3d software.

You’re overstating the simplicity of a scaling a GAN well.

GigaGAN is the best quality out of those and requires 7 loss functions and is incredibly complicated.

Sure GANs can scale, but Diffusion models are drastically easier to scale.

This wreaks of performative grandstanding.

Dictating the tools that artists use for a commission is punitive and moralizing.

Let the artists decide the morality of their own profession.

My definition would be that spiritual gateways are tools that help you have spiritual experience.

Plenty of people have had that shared effect from psychedelics.

I’m sure what baseless belief you see.

You admit that there is plenty of recreational and therapeutical use for them and then that leads you to the conclusion that they’re not spiritual gateways. I find this surprising as those are one in the same to me.

Spiritual experiences have been about recreation, understanding others, the self, personal growth, coming of age, etc…

I feel like your post in general is full of acknowledgment of these drugs capabilities followed by attempts to downplay it with pejoratives like “chemical glitching” or “lower dimension”. It’s quiet confusing that you seem to think positively of the drugs but then dunk on them in the same breath.

I don’t think being able to explain, categorize and describe phenomena needs to make them less significant or impact. The mechanism of action being non mystical doesn’t need to change the significance of the experience.

This reminds me of a quote from Margaret Boden about her work on understanding creativity in a academic context.

> A scientific understanding of creativity does not destroy our wonder at it, nor does it make creative ideas predictable. Demystification does not imply dehumanization.

It has depth as well. You just need to copy in the relevant docs.

It’s writing fault tolerant data scraping scripts for me I don’t know how to write myself.

It’s noble to want to help the poor but the economy isn’t zero sum in that way.

Almost all the surplus we have could be traced back to our discoveries in physics leading to machines.

People lived much much poorer lives without modern physics discoveries.

The expects return on physics is so great we should find ways to afford more long shot experiments not less.

> Since Pepe, authorities have been trying, unsuccessfully, to limit hippos' numbers by chemically sterilizing them or castrating them. This is a hard task given their immense size, aggression and internal testicles.

I’m not convinced yet, but if they can provide nice templating and reusability.

There is plenty room for code reuse in prompting.

Touché, that poem is a show stopper

If it only takes a simple sentence to set you off. You should be careful how much you’re reading into peoples language. The worlds not filled with facists in disguise. Look at people that way and you’ll see it everywhere. It will drive you mad.

There is no true scotsman. Doesn't mean we can't have truth

Supply of games is a big factor here as well.

Way more quality game makers

This isn't how large language models work. Deep Features are much more rich than this. It's not random the models have their own sense of taste, and you can easily control it with comments specifying what you care about in the code you are going to write.

I think the opportunity cost of better employee will make it hard for that to scale.

If you’re that good at scaling systems like that someone with legit traffic would be able to pay more than the spam companies.

Site is down.

Here's the google cache:

https://webcache.googleusercontent.com/search?q=cache:8E9sp-...

We've hit a couple of inflection point.

The numbers of researchers and research labs scaled up that there is now many well funded teams with experience.

Public tooling and collaboration has reached a point where research happen across the open internet between researchers at a pace that wasn't before possible. Common Crawl, stable diffusion, hugging-face, etc...)

All the techniques that took years in small labs to prove as viable are now getting scaled up across data and people in front of our eyes.