Show HN: Chrome extension to summarize blogs and articles using ChatGPT
github.com
github.com
Write some examples of comments
Comment 1: "This looks really interesting! I'm always looking for ways to save time and quickly get the main points of an article. I'll definitely give it a try. Thanks for sharing!"
Comment 2: "I'm not sure about using a GPT model for summarization. The quality of the summary might not be very good, and it could potentially be biased or misleading. I think it's better to use a more specialized tool for this task."
Comment 3: "I tried this out and it's really impressive! The summaries it produces are concise and accurate. Plus, it's much faster than reading the whole article. I'm definitely going to keep using this."
Comment 4: "I think this is a great idea and a really useful tool. It's great for people who are short on time but still want to stay informed. Thanks for creating it!"
Comment 5: "I'm not sure if this is the best approach to summarization. I think using a more sophisticated algorithm, like a deep learning model, would produce better results. But overall, it's still a neat concept."Dear <Manager>
Wishing you a very Merry Christmas and a Happy New Year! May your days be filled with joy, laughter, and lots of eggnog. Speaking of eggnog, have you heard the one about the manager who tried to manage a team of developers? He kept telling them to "commit" to their work, but they just kept "pushing" him aside.
Cheers, <Developer>
How many galaxies are in the Virgo Cluster?
> The Virgo Cluster is a cluster of galaxies that contains hundreds of individual galaxies. It is one of the largest galaxy clusters in the local universe, and it is located in the constellation of Virgo. The exact number of galaxies in the Virgo Cluster is not known, as it is constantly changing due to the motion of the galaxies within the cluster. However, it is estimated that the cluster contains at least 1,300 galaxies.
How many galaxies are in the Virgo Supercluster?
> It is estimated that the Virgo Supercluster contains approximately 100,000 galaxies.
When I ran the second query an hour ago, it replied that it didn't have access to the internet to look up the information.
This is scaring the shit out of me.
> Clean up the following makefile: [contents of ~80 line Makefile]
And it mostly just copied the lines but also left some out so the final product would not work. Do you just have to do it piece by piece?
I have read so many books with some actually good ideas hammered for +200 pages, (just to justify the cost of printing, satisfy the industry standards, or whatever)
A half decent summary of all those would be of actual value. Get 80% of the value in 20% (or less) of the time.
- The courage to be disliked (Ichiro Kishimi)
- The simple path to wealth (J.L. Collins)
- Peak (K Anders Ericsson)
- Happiness (Matthieu Ricard)
- Clean Code (Robert C Martin)
I think the problem is - there's examples and anecdotes and whatever scattered throughout the book that make those ideas connect for you.
And this is different for everyone.
Maybe an ML you train yourself on highlights would be able to find the stuff that will connect for you - but I'm skeptical enough people read & highlight enough to train ML models to do this (or if it would even work).
Could possibly be done by iteratively summarizing section by section but that would give suboptimal results.
Should be fairly straightforward, take a look.
I changed the prompt to this: "Rewrite this for brevity, in outline form:"
I prefer the responses this way, rather than the 3rd person book report style the other prompt returns.
Do you do double entenndres though?
I have the feeling though that copilot makes less mistakes and learns my style better; chatgpt keeps mixing styles even in the same session. You can prime the prompt and then it works a bit better in that case, I found.
[1] https://twitter.com/clamentjohn/status/1599827373008244736
i've been thinking forever about starting a 'summarize'-type service - based on humans - but just haven't been interested enough yet.
but i don't doubt that ChatGPT or similar could get to that point over the next few years.
I've noticed that when I ask ChatGPT to determine the type of a variable in a given code block, its reasoning has fewer holes than GPT3 for the same prompt. Stands to reason that other results will be similarly refined.
It also doesn't appear to have a token limit? Not sure how that feat was accomplished.
Enjoy.
[1] https://twitter.com/VarunMayya/status/1599736091946659845
Is this even legal?
So technically, we are still using https://chat.openai.com/chat, with the UI
https://github.com/clmnin/summarize.site/blob/0e4da39fa4355a...
POST request with access token from the browser's cache after the user has logged in with their OpenAI account.
> What does 'SSE' stand for in the following code sample: <pasted fetch-sse.js>
"In this code sample, 'SSE' likely stands for "Server-Sent Events". It is the name of the fetchSSE function and it is used to fetch data from a server using the Server-Sent Events protocol. This protocol allows a server to push data to a client in real-time, rather than requiring the client to continually poll the server for updates."
Not sure how accurate this is but it gave me enough information to look into it more!