going_ham

At this point, I won't hesitate a single bit of doubt that this company literally stole stable diffusion and built on top of it. Like how will we ever know? If they were so good they could have released before SDXL. But they waited.

Yes! That would be so cool. Also they are supporting Nvidia 2000 series GPUs which is completely bonkers!

This really put NVIDIA in clutch for the gaming.

NeRFs are cool technology that has its place and use. It is going to help with scene reconstruction and so on. And that is the reason why these CV researchers are flocking into using it. Despite knowing the limitations, they are giving their best to improve this technology. What I believe is NeRF is insufficient to be meaningful on its own. If you look into the architecture from this problem (the paper in the post), it clearly shows that they have some refinement phase going later on. A single RGB 2D to 3D model is such an ill-posed, we have to consider a lot of priors before diving into it.

There needs to be more foundational work in this field that can outperform or even improve the NeRF-based techniques. And the current herd mentality of researchers should be changed into exploring the alternatives. There is a reason why expensive automobile companies still rely on the physical modelling of their design. It's hard to simulate the physical conditions only through CAD modelling. Sure NeRFs are cool and they can make impressive results. That doesn't necessarily mean it is the means to an end. Look where rasterization brought us! NeRF is like rasterization. It is going to be used. But highly quality graphics was possible through GI and ray tracing! NeRF needs something equivalent that is physically grounded.

Now that is the same idea I also believe in. People are just capitalizing on it like it's a child's toy. This thing can have profound societal implications but no one writes about it. Is there a reason?

Also some of the people hyping it around are confident. How can they be so confident? Is it just blatantly lying into another's face and hoping you won't get caught?

Haha. That was a funny response from the GPT. I attacked your bot unintentionally!

On serious note, do you think as society progresses with use of AI, our brain will be geared toward filtering everything or we will lose the trust system that is prevalent in the society?

Think about it, the more people are being shut for bot, they more likely they will stop interacting online and this might eventually lead to a lot of people discarding the interaction. For most part, life is pretty average. And if average people are out, what might be the implications?

This was immediately my first thought when I saw the product. When is the expiry day? Why aren’t they making it clear?

Also I think this will inspire other companies and help those companies (cough microsoft, apple) create their own version which they will integrate to their own lineup. It will be interesting seeing it in future.

Disclaimer: I am not AI and I typed this response in my tablet. People for some reason have tendency to assume I answer like a bot.

Just some ramblings and experiences. There are tech related content too. I am thinking of updating and revamping it.

https://entangledlogs.com

My circle is 97.4% perfect, can you beat that? https://neal.fun/perfect-circle/

The dot is the center! Try to adjust to that!

Sorry doc, I wrote that comment in a smartphone without putting any thought. What I wanted to say was: > there are 7B parameters. A parameter is a weight assigned to single neuron.

I hope this clarifies the answer now.

Now that is done I am quite curious on how you came up with the idea it was written by ChatGPT? I just wanted to simplify as best as I could. It’s funny you thought it that way.

What could I have done so that it didn’t sound like response from ChatGPT? I am asking it to prevent future misunderstandings. I thought my grammatical errors would be enough to show it wasn’t a ChatGPT response.

Looking forward to your reply!

1. The trained model has 7B parameters or weights for each neuron.

2. It can handle upto 8k tokens. Tokens are usually some representation for a word. If your tokens are characters then, "h", "e", "y" represent 3 tokens for hey. Most of the algos use byte pair encoding. For example "hand-le" has two tokens "hand" and "le". This is a very crud example which is enough to give the gist but is not accurate. You can look into byte pair encoding for more details.

3. The token size 1.5T token means they have huge variations for input and output. Simply put, it was trained on large data corpus.

I hope this simplifies it. You can research further if you are interested! Hope it helps!

This is a great example to show the underlying mechanism!

You've also seen a bunch of my content on Hacker News because I'm one of the only people writing about it - if very much like not to be!

With all due respect, I would also like to market someone else who has also been posting similar content, but for some reason those posts never make it to the top. If you don't believe me, you can check the following submissions:

[0] https://news.ycombinator.com/item?id=35745457

[1] https://news.ycombinator.com/item?id=35915140

They have been consistently putting the risks of LLMs. Thanks for spreading the information though. Cheers.

What does this imply?

Not to sound rude, but how are you guys going to determine differences between user input and say, an input from an external sources like pdf, email, webpage, webapps? Do you have thoughts on it? If I make an application, I will want to link to external systems.

If there isn’t any way to distinguish it, I bet the attack surface is too large. If it is restricted to QA without external interface, then usability is also restricted. Any thoughts about it?

Isn't this similar to your idea? https://github.com/openai/openai-python/commit/75c90a71e88e4...

Correct me if I am wrong, but the way I understand is that, when LLMs have to process a certain text, every word will get tokenized into some vector representation. So, if you insert the new special token and wrap data around, it is not the fact that you can ignore the entire prompt. Because as soon as you have to prompt the model, you will be using the entire tokenized sentence. This would mean that even if there is a special token somewhere, the model will not be able to ignore the token before/after that special token. So what will happen to the model if somewhere there is prompt that overrides this special token?

I am using sofle v2. It costs less than 200!

Getting split keyboards. They are the best. The only regret: I didn't buy it sooner.

I am pretty sure you hit the spot. I do not even write notes these days. I just work and if it is important enough for me, I trust my brain to recall. If it can't recall, I search the web. However, this time I have different context so each new search is unique. I discover different ways of doing things. Sometimes, it's better than what I used to do. As a result, I only focus on high level problem solving. This way, I don't care about small details and remain faithful to general context. What do I gain from this? Long term ability to solve problem. Each problem I solve, makes me think more on my thinking process. As a result, I optimized surroundings in a way it helps me think and reason rather than note and forget.

More like a generalist who also happened to be one of the greatest. You can read his book "Surely You're Joking, Mr. Feynman!" It adds some characteristic to his flamboyant personality.

I have similar opinion! Many people are not taking into account the fact that this learning technique significantly outscales the average in terms of numbers. While the very best may not be affected, the fact that it generates results that are on average or slightly below average is concerning. Additionally, its ability to imagine and contextually change random garbage means that there is already room for improvement. Given the compute and resources available to top labs, they will likely be able to add more context, which could exacerbate the problem.

To my knowledge, no other system has been able to retain context, imagine garbage, and refine it based on the given input context. For details, it can imagine a game, change the premise of the game and craft stories or lines based on the new context. It doesn't even have to be perfect. It is already capable of doing 60% job. Wait till it reaches 80%.

Now you do not have one skill, but you can have multitude of average looking skill on top of your own craft. Further, this thing provides a good entry point to dive into research and investigate. Just like how stable diffusion sparked curiosity on art culture, technicalities, photography, this can do it but on wider context and problems. Therefore, it is not the matter of will it, it is just the matter of when?

I like this grammar hack.

```

grammar = """

Response ::= "Alice\n" (ListOfCmds | SingleNaturalLangResponse)

ListOfCmds ::= (BashCommand\n)+

BashCommand ::= "$ " AnyString

SingleNaturalLangResponse ::= AnyString without $"""

```

Dear executives,

You have AI that is capable of solving your interview questions. These questions were designed to select the best candidates. Now you have AI that is cheaper and faster. Please, fire your engineers and embrace the next generation of programming. Why would you need engineer when a bot will do it for you? After all, your candidates suck at leetcode interview.

when you're not a genius and you do things for the fun of it it's very probable you'll end up wasting your time. This is why usually you have advisors that help you to find important problems that need to be solved.

Also add if there are not great examples to learn from. Most radical ideas require mastery of the topic. Incremental work can be done copying and simply adding one on top. Advisors have gone through all of this experience. Most people underestimate power of good mentors.

This scares the shit out of me, when my career hasn't even started yet!

It seems 2020-2022 was boom for tech industry. Now the rate is going backing to pre-2020 which shows it was already in decline! That is quite surprising.

which country are you from? Just curious..

If you can formulate your problem in the sense of classical planning problem, there are different sets of planners that can solve the problem. These methods are explainable. Some of them even give the optimal solution to your problem if you use optimal heuristics. This is what he means by that statement. The only issue is, right now the classical planning is hyperfocused into few problems and most of the methods work on narrow domains like path finding, logistics, constraints based problems and so on.

Why is this response being downvoted? It seems to clearly reflect the hiring bias present in the industry. Please upvote it so that people can know how to overcome the inherent bias on quality.

There is some serious flaw with software people (all stakeholders). It's like they are addicted to "staying relevant." Most of the time people don't need new features. Why isn't it being drilled over to the project managers? Instead of maintaining and supporting what they created, they are more into ruining and adding bloats of features. (looking at you, Jira).

There are enough engineers to have custom tailored softwares. Figma was a stellar example of one well-designed single purpose software. All they did was maintained and optimized the crap out of it!

Game dev is one crazy field that proves how bloated the modern tech stack is.