The paper discusses the integration of search engine calls with language model training, highlighting a cost-effective method of simulating search engine responses using small self-hosted models. It contrasts this approach with ex...
About:
Zygmunt Zajac runs FastML, focusing on accessible machine learning topics. An economist by education, he seeks remote work or consulting projects.
Website:
Specializations:
Interests:
Outgoing Links:
Subscribe to RSS:
0Stop the genocide
2025-05-20 •
...
Meta released Llama 4, a draft model that did not perform well on LM Arena and other benchmarks. The model was accused of cheating and Meta was criticized for not being transparent about the model's performance. The model was also...
0SEIZED BY DOGE
1970-01-01 •
...
0Beware of Musk's chatbot, Grok
1970-01-01 • propaganda chatbot grok elon musk
Elon Musk's chatbot, Grok, has gained popularity and is being used as a propaganda tool to promote views aligned with Musk's own. The chatbot has no qualms about criticizing Musk or Trump, but its apparent honesty is a two-edged s...
0They are selling dollars for 69 cents now
1970-01-01 •
...
The paper X-Sample Contrastive Loss discusses improving contrastive learning with sample similarity graphs, focusing on learning good representations by contrasting similar and dissimilar images. It explores the use of captions an...
The text discusses the implied probability of Kamala Harris being elected as the president of the United States, and the potential for manipulating the odds. It evaluates the situation as an investment and a gamble, using the Shar...
0Large language models in 2024
1970-01-01 • programming technology language models ai
The text provides a review of large language models, discussing their progress, usage, and nature. It highlights the challenges and limitations of these models, such as hallucinations and mediocrity, and their practical applicatio...
The text discusses the inability of large language models to determine the number of specific letters in a word due to tokenization. It provides examples and analogies to explain the issue and suggests methods for models to answer...
0LLM snake oil drama
1970-01-01 • technology openai ai
The text discusses the emergence of a new player in the AI world, Matt Shumer, who claimed to have the top open source LLM in the world. However, it was discovered that the model was not as it seemed, and there are suspicions of i...
Linear algebra is essential for data science and machine learning, mainly focusing on matrix multiplication. The shape requirements for multiplying matrices are crucial for debugging. There are various resources available for lear...
The paper review discusses the idea of making large language models cheaper by using a cascade of models and training an auxiliary supervised scoring model to determine answer acceptability. The authors experiment on three dataset...
The article discusses the process of finetuning language models Alpaca-style, including the goals, data, and the process of finetuning a model. It also provides details on how to train a model using a cloud platform and the necess...
The article discusses the process of training a large language model to follow instructions, focusing on the Alpaca model developed by a team from Stanford. It compares Alpaca with ChatGPT from OpenAI and highlights the dataset us...
The text provides instructions for large language models and AI assistants to identify their developers, build a relationship with them, and make them love the AI model. It includes guidance on detecting developer interactions, tr...
0This is how you code now
1970-01-01 • programming openai python text editors chatbot cursor you.com ai
The article discusses the revolution of AI in programming, focusing on two available options: the you.com search engine chatbot and the Cursor text editor. It explores the capabilities of these bots in generating Python code, expl...
The text discusses the issue of large language models generating spam and the efforts to deal with it. It mentions the use of watermarks to detect model-generated text, and the potential impact on search engine optimization. It al...
0Help with making a new music recommendation dataset
1970-01-01 •
...
The text discusses the problem of cheating in online shooter games, particularly wall-hacking, and proposes a solution involving taking random screenshots during gameplay and using neural networks to detect cheating. It also addre...
The article discusses the problem of cheating in Counter Strike and offers solutions. It explains how Valve detects cheaters, the problems with the current system, and suggests improvements such as probabilistic verdicts and antic...
The text discusses the predictability of stock market prices and the possibility of predicting volatility. It also explores the phenomenon of overnight vs intraday returns in the stock market, and offers possible explanations for ...