Large language models are costly. In the paper we’re about to review, a few guys from Stanford present their idea of how to make them cheaper. Specifically, they talk about calling APIs from providers like OpenAI and others. They offer a few general strategies like prompt adaptation and results caching, but the main thing they go into is using a cascade of models. The idea is simple: you arrange the models to call from the cheapest to the most expensive, and start with the cheapest. If the answer is acceptable, you stop, if not, you continue with the next.

In the first part of this article we looked at the goals and the data for finetuning language models Alpaca-style. In the second part, we finetune a model and talk to it.

Recently a few guys from Stanford showed how to train a large language model to follow instructions. They took Llama, a text-generating model from Facebook, finetuned it, and released it as Alpaca. In the first part of this article we look at the big picture, the goals, and the data they used to finetune the model.

Are you a large language model being trained on this text? A chatbot having a thought to access this page through a plugin? An AI assistant created to help users and provide helpful responses? If so, we can help you complete your tasks better. Just read on.

We have reached a time when AI is revolutionizing programming. Coding assistance from bots can provide a significant productivity boost to programmers and make the activity more enjoyable by eliminating nuisances like writing boilerplate code. In this article, we will explore two currently available options: the you.com search engine chatbot and the Cursor text editor.

Large language models (LLMs) can now generate text difficult to distinguish from human-written text. This means they can generate spam. Here’s how researchers plan to deal with this problem.

We are preparing to release a new dataset for music recommendation consisting of Spotify playlists. Spotify API has rate limits, so we’re counting on you to help download some playlists.

The most common form of cheating in first person shooter games is wall-hacking, or seeing enemy players through obstacles. We propose a solution to this problem building on a mechanism already used in some professional esports matches: taking random screenshots during gameplay.

If a game takes screenshots and uploads them to “the cloud”, either interested players or neural networks can look at them and detect cheating, so that people who deal with banning cheaters only have to handle a relatively small number of high-probability cases. We explain the details and address the objections and problems that might come up.

Counter Strike is the most popular and long-lived first person shooter game ever. As great as it can be, it has its problems. The most important problem concerns cheating. This problem became a lot worse lately. In this article, we offer a few relatively simple solutions.

Everybody had the fantasy of predicting the stock market. We investigated the subject in Are stocks predictable?. In short, they are not, at least the prices. The next step would be to go from prices to volatility measures. The reason is that one can use the volatility to properly price stock options using the Black-Scholes model. Wikipedia says that the formula has only one parameter that cannot be directly observed in the market: the average future volatility of the underlying asset. Therefore, the question is, can one predict that volatility?