muglug

15 ms·

211.
by muglug 2y ago
This exactly. If Microsoft had created its own fine-tuned-MSFT-data LLM and seen vastly better results on internal tasks, then they’d be publishing papers about it, and also packaging that up & selling to customers.
212.
by muglug 2y ago
There’s significant money to be lost (lawsuits, massive reputations damage) if it was discovered they weee deliberately training on data they had guaranteed that they were not training on.
213.
by muglug 2y ago
It’s... not? Even for the low bar of a geek joke it makes no sense since GPT-4 is an LLM.
214.
by muglug 2y ago
Edited to specify I'm talking about programming languages...
215.
by muglug 2y ago
Passively watching YouTube is probably not the right way to learn any progamming language.
216.
Re-architecting Slack for our biggest customers(slack.engineering)
2 points by muglug 2y ago|0 comments
217.
by muglug 2y ago
> What's the need to taint the impressive result with apples-to-oranges comparison? Most of DeepMind’s research is a cost-centre for the company. These press releases help justify the continued investment both to investors and to th
218.
by muglug 2y ago
Following a long trend started by Magic Leap, the videos are entirely made up. I bet the display is very faint outdoors, just like the AR headsets that came before.
219.
by muglug 2y ago
“[it] apparently did not undergo adequate quality checks” You don’t say
220.
Quadratic Approximation of Cubic Curves (2020) [pdf](ttnghia.github.io)
2 points by muglug 2y ago|0 comments
221.
by muglug 2y ago
John Carreyou, writer of this article, exposed Theranos while at the WSJ.
222.
by muglug 2y ago
This was my attitude when I stopped maintaining a large open-source project that I had created. None of us last forever, in life or even just in this industry. To have brought about some sort of positive change is more than good enough.
223.
by muglug 2y ago
The creation of computer science degree courses did not push down the price of software engineers.
224.
by muglug 2y ago
I think this page is a few years old
225.
by muglug 2y ago
I’m sure you’re a great parent, but you’re not a convincing advocate for the time-saving power of LLMs. Either way I was wrong — this project does look like a cool idea, and I’m sorry for dismissing it entirely.
226.
by muglug 2y ago
I don't think you're a bad parent, but I do think you were using hyperbole when describing how much time ChatGPT saved you. I also don't believe that you would spend a week away from your kids to make a 1,200-line app that ju
227.
by muglug 2y ago
It was a glib comment, but I fundamentally reject the premise that we'll have more personal time because of ChatGPT. It's like saying that buying a faster computer that compiles quicker means you can spend more time with your fami
228.
by muglug 2y ago
So three years ago you'd spend less time with your kids in order to write a 2,000-line FOSS app for understanding Bash scripts? I'm not sure that's a good thing.
229.
by muglug 2y ago
Stuff can creep into people’s lives while still being unprofitable for companies that are building on it. Crypto is clearly a big deal to a bunch of folks, but when large tech companies jumped on the blockchain bandwagon almost none of thos
230.
by muglug 2y ago
Maybe they think it’ll get printed out for CEOs to read? Love your blog!
231.
by muglug 2y ago
A whole lot of large tech companies are proudly shipping their “it demoed well” hackathon projects. Investors are far more eager for these things to get released than customers are to use them.
232.
by muglug 2y ago
Unlike many other AI-themed papers at Meta this one omits any mention of the model output getting used at Instagram, Facebook or Meta. Research is great! But doesn't seem all that actionable today.
233.
Major record labels file lawsuits against Suno [pdf](s3.documentcloud.org)
2 points by muglug 2y ago|1 comments
234.
by muglug 2y ago
Some buggy JS — it only reports one city's temperature transition numbers, then those same numbers are repeated for each selected city.
235.
by muglug 2y ago
For people unfamiliar with Enzyme and RTL, this was the basic problem: Each test made assertions about a rendered DOM from a given React component. Enzyme’s API allowed you to query a snippet of rendered DOM using a traditional selector e.g
236.
by muglug 2y ago
Yeah — a bunch of inspirations for this. Wouldn't be surprised if the designer was inspired by newly-released Emigre typefaces like Democratica.
237.
by muglug 2y ago
Way too many differences for it to be a feasible origin typeface. This post is based on a few similar-looking letters but they’re not even that similar-looking: the O, for example, has a different curvature.
238.
by muglug 2y ago
> The college student studying engineering met Musk in the early 2010s during her summer internship at SpaceX. Musk and the woman went out for a meal after she sent him ideas about how to improve the company, she told friends. They bonde
239.
by muglug 2y ago
If the tasks were public then it would be trivial to have a human figure out the answers, and then to train an LLM to memorise those answers.
240.
by muglug 2y ago
I think you're wrong about that — I just tried prompting ChatGPT 4o to show all its working before giving an answer. It was still incorrect, but when asked to show its working it formatted the answer prettily.

More ›