Member-only story
How AI Summaries Supercharged My Reading
Modern models are so good at cutting through noise that they’re revolutionizing how I consume information, helping me focus on what’s truly important
I recently read a comment expressing skepticism in using AI to summarize text. Now this is not the first time I’ve seen people criticize AI summarization, but the previous time I saw a comment like this was a long time ago. Way back when AI was not very good.
I remember when ChatGPT first came out I tried to summarize articles with it and it was pretty terrible. I distinctly recall once that ChatGPT only summarized half of the article. I think it was a problem with the amount of text in the context, it simply forgot about the second half of the prompt.
But things have improved a lot. GPT-3.5 Turbo, the model originally powering ChatGPT, supported 4k input tokens at launch. Modern models support 100k to 2 million input tokens and are way better at processing that many input tokens.
Google announced they were able to achieve over 99.7% on the Needle in the Haystack benchmark (a benchmark measuring the ability to identify obscure data in long context) at 1 million tokens. And this was Gemini 1.5 Pro released a year ago. It would not surprise me if things…
