menzoic

"During its thinking session it reached the conclusion that this task is not feasible in one shot. It then stopped and explained that to me."

I've seen this happen with GPT-4 with zero shot prompts. Similar to the author "negotiating" allowed it to continue with an iterative approach.

They are. The wealthy take low interest loans out on their assets to invest in more assets

This has nothing to do with earning a livable wage. It’s not like the top 2% are only earning enough to barely get by.

The article doesn’t explain how it works. It only lightly covers what was disproven

Also title is wrong, this isn’t a Data Science conjecture. Data Structure or Computer Science is what it is.

Having memory is fine but choosing the relevant parts requires intelligence

I’m also surprised that a slight change in C1GalT1 expression has such a big effect on the glycocalyx. That enzyme is notoriously high-capacity, so you’d think it runs at near-saturating levels most of the time. My hunch is that something else in the pathway is hitting a bottleneck, possibly a chaperone like Cosmc that’s needed for proper enzyme folding, or maybe there’s some substrate competition with parallel glycosylation pathways. Once that bottleneck is reached, even a small shift can have an outsized impact on final glycan structures. In mice specifically, there might be unique regulatory quirks that amplify this effect, and you end up with a dramatic change in the mucins even though the enzyme itself seems too powerful to be the limiting step.

I’m building what you thought it was. DM to beta test!

> The fact is that “Probabilistic plagiarism” is a mechanical process, so as much as you might like to anthropomorphize it for the sake of your argument

I did not anthropomorphize anything. “Learning” is the proper term. It takes input and applies it intelligently to future tasks. Machines can learn, machine learning has been around for decades. Learning doesn’t require biology.

My statement is that it is not plagiarism in any form. There is no claim that the content was originally authored by the LLM.

An LLM can learn from a textbook and teach the content, and it will do so without plagiarism. Just as a human can learn from a textbook and teach. Making an analogy to a human doesn’t require anthropomorphism.

> reproduce books word by word, page by page

This statement is a figment of the commenters imagination with no basis in reality. All they would have to do is try it to realize they just spouted a lie.

At most LLMs can produce partial excerpts.

LLMs don’t store the data that it’s trained on. That would be infeasible, the models would be too large. Instead, it stores semantic representations which often uses entirely different words and sentence structures than the source content. And of course most of the data is lost entirely during this lossy compression.

We have to understand what plagiarism is if making claims of it. Claiming that you authored content and reciting content are different things. Reciting content isn’t plagiarism. Claiming you are the author of content that you didn’t author is plagiarism.

> it's perfectly capable of publishing protected content

At most it can produce partial excerpts.

LLMs don’t store the data that it’s trained on. That would be infeasible, the models would be too large. Instead, it stores semantic representations which often uses entirely different words and sentence structures than the source content. And of course most of the data is lost entirely during this lossy compression.

For one Sam scraped under the veil of a corporation which helps reduce or remove personal liability.

Second, if the crime was the act of scraping then it’s directly comparable. But if the crime is publishing the data for free, that’s quite different from training AI to learn from the data while not being able to reproduce the exact content.

“Probabilistic plagiarism” is not what’s happening or even aligned with the definition of plagiarism (which matters if we’re talking about legal consequences). What’s happening is that it’s learning patterns from the content that it can apply to future tasks.

If a human reads all that content then gets asked a question about a paper, they too would imperfectly recant what they learned.

>I wouldn't want everyone tracking my airtags.

That’s kind of how the whole system works.

TIL you can learn super powers on the internet

This is literally a superpower

What a strange article. In reality it’s about what they don’t do not “can’t” do.

This type of analysis is heavily rooted in personal preference. Critics have their own biased. Relying on mass review system should represent more of what the average person like rather than some elitist idea of a critic of what should be.

Learning how to use ̶C̶a̶l̶c̶u̶l̶a̶t̶o̶r̶s̶ LLMS is probably the skill we should be focusing on.

> I don't think Netflix and Uber had even a fraction of the competition that this field will have.

Uber had significantly more competition than all of these companies combined. Including from China which they were forced out of.

LLMs will continue to be used by increasingly more developers

Is this a joke? Can’t tell. OpenAI uses PyTorch to build LLMs

How does Product Chart use LLMs?

Autocompleting secrets sounds unsafe.

> The receiving side offers tab-completion on the codewords, so usually only a few characters must be typed.

How do they know if you read it or not?

Clearly the fault of a terribly lead engineering organization. Mistakes are almost guaranteed to happen. This is why good engineering orgs have guardrails in place. There were no guardrails whatsoever here. Accounts used to manually access adhoc production databases should not have delete permissions for critical data. And worst off all no backups.

GPT-4 is the best coding model by far and you can’t fine tune it yet. Start with GPT-4 and RAG.

Fine tune is good for changing behavior. For example you could create a dataset that makes the model directly output fully formed projects. Using RAG alone would require a bunch of prompt engineering to get the same.

The electronic flow meter analogy is killing me

This isn't unique to Google. This is encouraged at Uber and Airbnb as well. I would think its a common practice. The idea is to show that you consider alternatives and did due diligence. Following the way its always been done isn't necessarily a good thing. If its a different product area there could be better approaches unique to that area. Its not fake work, it just proves you actually considered other approaches instead of blindly following status quo.

Or complain so that things can change

> But instead of changing their name and their mission, and returning the donations they took from these wealthy tech founders, they used the benevolent appearance of their non-profit status and their name to mislead everyone about their intentions.

Why would they change their mission? If achieving the mission requires money then they should figure out how to get money. Non-profit doesn't actually mean that the corporation isn't allowed to make profit.

Why change the name? They never agreed to open source everything, the stated mission was to make sure AGI benefits all of humanity.

>wouldn't there be problems with the speed and cost in electricity

proof of stake blockchains don't have those issues