HNHacker News
TopNewBestAskShowJobs

wskish

277 karma · joined September 18, 2014

Technical founder (Ruckus Wireless and Cogniac); kicking ideas and playing around towards #3.

See also @wskish on twitter.

submissionscomments
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
For anyone interested in seeing summaries without telegram you can now view the summaries for the corresponding front page here:

https://news.jiggy.ai

wskish··on Bye, Twitter
I have a lot of respect for Tim Bray as a technologist and have followed him closely over the years. I have also been thinking about Paul Graham's "The Four Quadrants of Conformism" a lot lately. I am curious to hear where other folks who are familiar with both would perceive Tim mapping into this space?
wskish··on The Need to Read
I tweaked the GPT prompt to include the domain name from the story url as additional context. That fixes this one and a few other problematic ones. Here is a typical summary of this story now:

"This is an essay written by Paul Graham in November of 2022. In it, he talks about how reading is necessary not just for acquiring knowledge, but for learning to write well. He argues that there is no substitute for reading, and that people who want to have ideas need to be good at reading."

wskish··on The Need to Read
We prepend the prompt prefix to the page text to create the actual model prompt. GPT doesn't have any such abstraction.
wskish··on The Need to Read
we are strategically prompting it as such: https://news.ycombinator.com/item?id=33750013

open to all ideas here

wskish··on The Need to Read
Here is the output of GPT reading and summarizing this (see https://news.ycombinator.com/item?id=33748363). Model seems to be hallucinating the author. Any thoughts on how to improve it?

"The Need to Read is an essay written by Kevin Kelly. In it, he argues that reading is essential not just for acquiring knowledge, but for learning how to write. He cites the example of a science fiction book in which reading has been replaced by a more efficient method of knowledge transfer. He argues that even if such a method existed, it would be insufficient because it would not allow for the kind of discovery that writing does. He concludes that people who want to have ideas can't afford not to read."

wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
Here are a few recent ouputs from the service:

Editing Binaries in DOS (2002) https://news.ycombinator.com/item?id=33737207 This is a technical report on editing binaries in DOS. The author describes how to use the debugger program DEBUG.EXE to edit binary files without requiring additional tools. The author provides two examples of how to edit a binary file. The first example edits a string of bytes in an executable file. The second example edits machine instructions to alter the behaviour of the program.

Elon Says He’ll Make His Own Phone If Apple and Google Deplatform Twitter https://news.ycombinator.com/item?id=33750616 This is a news article discussing a potential new phone by Elon Musk. The article discusses the controversy around whether or not Twitter should be banned from Apple and Google's app stores. It also discusses the possibility that Musk may make his own phone if Twitter is banned.

IRS warns taxpayers about new $600 threshold for third-party payment reporting https://news.ycombinator.com/item?id=33750598 This is a news article from CNBC discussing the IRS's new $600 threshold for receiving Form 1099-K for third-party payments. The article explains that the change applies to payments from third-party networks, such as Venmo or PayPal, for transactions such as part-time work, side jobs or selling goods. The article also explains that the IRS is reminding taxpayers that they may receive Form 1099-K for transactions they don't expect, such as reselling Taylor Swift tickets at a profit, for example. The article advises taxpayers that if they receive Form 1099-K for personal transactions, they should contact the issuer for a correction.

wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
haha yeah thats pretty funny. We need to figure out how to derate these models when the have no actual clue. And also how to scrape the source content even better...
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
yeah news articles really tend to highlight the "controversial" aspect. Unfortunately they are also the hardest ones to scrape...
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
yeah, one of the biggest issues on HN is that folks comment without reading the story article. Perhaps this sort of summary of the story article can help drive a higher quality engagement.
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
Definitely, it would be great to figure out how to summarize the discussions and work them into the UX. Anyone please send PRs with ideas around how this might work!
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
wow, good stuff; would be super entertaining to try to duplicate that with models
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
We use the GPT2 tokenizer (which is apparently the same as the GPT3 tokenizer) to count the number of tokens in the extracted input text and brutally truncate the input text to keep it under the limit. There is a lot of room for more finesse in this area if anyone wants to help out. See open issues in github.
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
It is perhaps unnecessary, and I don't have any data that would support that it makes things better, but I am basically guessing that in the training set of {mostly the entire internet} the responses to questions that include "please" might be ever so slightly higher quality! Perhaps it is just a superstition...
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
what was n-gate? I have a pet theory that humor will be the actual litmus test for AGI.
wskish··on Show HN: Open Source Bot That Summarizes Top Hacker News Stories Using GPT-3
FYI, here is the current prompt prefix we prepend to the story text before sending to GPT-3. Any suggested alternatives? Will try them based on points here.

"Provide a detailed summary of the following web page, including what type of content it is (e.g. news article, essay, technical report, blog post, product documentation, content marketing, etc). If there is anything controversial please highlight the controversy. If there is something unique or clever, please highlight that as well:"

wskish··on Twitter at 1am on a Friday
>> shared suffering cements lifelong bonds. > I'm sorry, is this a cult or a workplace? Why is this a good thing?

There are many contexts where some amount of suffering is a prerequisite for achieving great things. For example, from personal experience, both startups and mountaineering require some amount of suffering to achieve great things. The bonds formed from shared suffering are like top 25th percentile.

wskish··on Twitter at 1am on a Friday
can you share the datapoints you are referencing that lead you to this conclusion? My experience over 30 years spent in many startups as well as other domains such as mountaineering where shared suffering cements strong bonds leads me to a different conclusion. Perhaps we just differ on our assessment of EM's motivations.
wskish··on Twitter at 1am on a Friday
given the strength of the network effects a subscription model is worth exploring, which seems to be what he is pursuing.
wskish··on Twitter at 1am on a Friday
people who share intense experiences and endure hardship together will tend to be very open to communicating again in the future. This is based on my past experience, not fantasy.
wskish··on Twitter at 1am on a Friday
I didn't quite follow, what are folks giving into?
wskish··on Twitter at 1am on a Friday
i guess we all have to judge the value of such human connections
wskish··on Twitter at 1am on a Friday
Of all those, "advertising" is the only fundamentally evil one, and he is chipping away at that one.
wskish··on Twitter at 1am on a Friday
I agree with you in ideal circumstances.
wskish··on Twitter at 1am on a Friday
historically, yes; but might that be a function of both the platform incentives (advertising, engagement) and poor technology?
wskish··on Twitter at 1am on a Friday
> what about his actions leads you to believe he's more capable of that than the thousands of people who were already working on it?

My sense is that the 1000's of people who where already working on it have been working on it from the perspective of how to make a billion dollars while working at it. EM is free from that constraint.

wskish··on Twitter at 1am on a Friday
I am curious about what you think could be the most relevant differences.
wskish··on Twitter at 1am on a Friday
I like to turn it around an ask what opportunity I would show up for at 1am.
wskish··on Twitter at 1am on a Friday
sorry I dont quite understand what you mean by "acceptable one" and "give them one". Which one are you referring to?
wskish··on Twitter at 1am on a Friday
crazy situation; all the more reason to draw a map to see where we are all at.
← PreviousPage 3 of 4Next →