One of the entire value props for Jev is that is exactly how it is supposed to work. That is one of the big claimed advantages over regular LLMs.
3,503 karma · joined July 18, 2019
The US is on the hook for its decisions. But Europe doesn’t get off the hook for its own.
Classifiers like classical NNs require:
- annotated data, potentially a lot of it - training - inference
#2 and #3 aren’t a big deal if you have an ML engineer, but #1 will always be a potential headache no matter who you are. The tradeoff is that they could be quite fast, cheap, and you can get probabilities, not just classes.
With LLMs you get:
- zero shot classification (no dataset or training required) - potentially can use third party model providers like OpenAI off the shelf. Don’t even need to host your own model.
The downside to LLMs is that they are comparatively slow and expensive to traditional classifiers. Historically they also were prone to hallucination or malformed responses, though not as much these days. You also can technically get log-probs back, but these aren’t equivalent to the classifier probabilities.
Jev gets you the zero-shot, zero-infra benefits of LLMs, while being closer to the speed and cost of traditional ML classifiers, as well as both classification and probability responses.
Early in my career I worked for a pharmaceutical giant as a statistician. One of the things we would work on was trying to determine how much the various advertising efforts the company did actually moved the needle. Every year was the same miserable experience. The company would spend millions and millions of dollars in various advertising channels. We would attempt to determine how effective it was, but the wayin which they advertised and the way we collected data made it very difficult to generate any kind of insight at all. We would make recommendations as to how to adjust execution moving forward to ensure we could reliably determine how much value we were getting out of these campaigns, all of which would be ignored. They would massage our findings to tell whatever story they needed and would do the same thing the next year.
The marketing division is its own organization, with their own incentives that don’t necessarily align with the broader company’s incentives. They also were heavily addicted to relying on external analytical consultants who could pump out all the slide decks and pie charts they wanted, which would inevitably show that the marketing was not only effective, but should be invested in more the following year.
For what it’s worth, as an American, I agree with them. The USA is proving to be unreliable, if not outright malicious, to our long term allies.
I don’t think truth aligns with either extreme. Both are approaches that have their place. But if it was an “uncreative” with no ideas that simply took the time to look and discovered that the three point shot was absurdly more efficient than basically any other shot in basketball, what does that say about all the “creative” coaching geniuses who failed to discover it in half a century? And if basketball is now “less creative” because of it, that’s an indictment of the fact that the league refuses to address the underlying issue.
I think it’s telling that no one bemoans the fact that we design our medicine using data driven methods. I’m sure it would spice up our lives if the pharma companies dropped the boring data analysis and statistical methods and just slapped some compounds together through free-flowing creative passion!
Both are based on pre-existing material but GoT source material predates the Walking Dead by 7 years.
I’m also not sure I understand the “ My place used pylint, flake8, black, ruff -- with hundreds of commits on every change”. These tools don’t add extra commits. You run them prior to your PR, get them aligned with the linting, and then commit the change you were already going to make. Thats 0 extra commits.
For someone ultimately arguing that there is too much effort spent on people’s coding styles, you are spending a lot of effort arguing about people’s coding styles. These tools are some of the most set it and forget it things around.
Therein lies the entire issue. Intellectual property is a paradox and a farce, but our current systems require IP as a concept to exist. 99% of human history existed without the concept of IP law, and yet people still innovated, learned, and shared knowledge. But now we are at a point where human beings can only survive if they have something they can productize and sell to others, and so for knowledge to be produced, it has to turned into something that is artificially scarce, so that those producing it can be compensated for their labor. Maybe this is unavoidable for this point of our evolution. I doubt it, as 99% of human history managed without the concept of IP, and we still produced and shared knowledge.
But even if it is necessary for this stage, it is still a dystopian model. And I imagine we aren’t far off from a point where we can provide for everyone’s basic needs, and therefore we don’t need to introduce artificial scarcity to make knowledge work economically viable.
In the case of AI, I don’t think LLM providers are required to make their models completely open source just because they were trained on the data freely available on the internet. Those model weights aren’t the original data anyways, so they aren’t even selling access to that original data. However, there is should be nothing preventing others from creating and selling something produced from those LLMs. Distillation, for example, is free game.
In short, you should be required to make your knowledge freely available. But once someone has access to that knowledge, you can’t stop them from using it however they wish, including if they wish to sell it themselves.
Second of all, “knowledge is free” doesn’t mean it is free to generate. It means that it literally cannot be owned. Property implies scarcity. If I use a thing, it makes it so you can’t use that thing. This does not apply to something like knowledge. The fact that I use your idea doesn’t mean you can no longer make use of that idea. Your analogy is flawed, because drinking a beer means that is a beer that cannot be used by someone else. Intellectual property is an oxymoron. There is nothing actually stopping someone from making use of someone else’s knowledge, and therefore we must artificially construct barriers in order to maximize the ability to profit off of it. In the most charitable sense, this in theory provides financial incentive for people to generate knowledge. In a less charitable sense, it is literal though policing for the sole purpose of extracting money.
I get that making use of knowledge without compensating the person who produced it is “theft” under our system. The problem is not with the act. It is with the system in which knowledge work can only support an individual if these dystopian measures are enforced. Frankly, most knowledge work is built on mountains of previous knowledge work that never is compensated for it anyways, so the model of the world isn’t even internally consistent.
Requiring a disclaimer is essentially admitting the content isn’t meaningfully different than human generated content. At that point, who cares? Just engage with the premise on its own merits, rather than on how it was written.
At the end of the day, the ideas within the content are what matters. An idea has or does not have merit regardless of if it was produced entirely by a person, or by a person using AI as an editor, or 100% generated by AI. If you need a disclosure on if an idea was produced by AI, you are saying that you have no interest on debating the content on the grounds of the arguments it is making, while simultaneously ceding you can’t tell the difference between someone using AI and someone who isn’t (which undermines one of the primary arguments against AI, that it makes for inferior outputs).
Me. In my limited experience, the more leaders are involved, the less likely the project is to be successful. The ideal situation is having leadership that is able to set a decent strategy, and then get out of the way of their subordinates and let them execute. Bad leadership micromanages and inserts themselves in the way of execution.
The status and “power” involved in setting strategy isn’t nearly as significant as the amount involved in micromanagement.