bfung

*InverseCramerGPT is what you’d want chuckle

stop running servers computing hashes that don't contribute to society? btc cough /s

I used to work for a tech company that used big data to convince large swaths of people to become energy efficient. One of the company metrics was how much energy was saved, accounting for our footprint as well; it was fun and gratifying when we crossed the saved more than 1TWh threshold, which also reduced the need for fossil fuel peaker plants. So these companies do exist, but the tech isn't sexy nor complex/hard/challenging ~ way more of a business and social component to these systems, and tech is only a small tool here.

Said company was acquired by Oracle and still running and doing well!

* https://www.oracle.com/industries/utilities/products/opower-...

* https://www.cnbc.com/2022/07/14/how-opower-sold-to-oracle-ha...

I’ve been on and off mac’s since the late 90s. Contrary to the article and similar to comments here, their OS was never that good until last 10 years - not the other way around.

As I got older, I got less time to fiddle with drivers on linux (Nvidia/noveau drivers ptsd, wifi drivers…) why can’t it just work!?

The title should be the other way around: Apple is making Linux Great Again.

Having limited but excellent hardware configuration makes writing stable software easier.

> There is an expectation of profits from the investment

Not everyone cares to 1) sell their house 2) at a profit. Some people just want to live in it and even die in it.

I love how the first reply to this tweet thread is Grady Booch, co-inventor of UML and pretty influential Software Eng. person in the early 2000s.

Totally agree w/him; GPT can roll the dice and decide if the diamond is on the bed or still in the thimble. It doesn’t prove it has a model of the world.

https://en.m.wikipedia.org/wiki/Grady_Booch

Not quite “hijacked” as the title implies.

The corp already was admin of the GitHub org, with contributors having a level of access.

Said contributors had their access lowered w/communication.

I say fork it and move on w/your lives. The project is mostly just a list/static webpage. Call the new one “free-apis-list” or something.

You are correct. The process of generating high-quality training data is a critical component of building and training AI language models like me. While much of the training data can be collected through web crawlers and other automated means, there is still a need for human-generated data to supplement and refine the training corpus.

There are several methods for generating human-labeled data, including crowdsourcing, annotation, and manual labeling. Crowdsourcing platforms like Amazon Mechanical Turk can be used to solicit human input on specific tasks, such as labeling the sentiment of a particular text or image. Annotation tools, such as Prodigy or Labelbox, allow data scientists to create custom labeling schemes and train human annotators to label data with high accuracy and consistency.

Manual labeling, while more time-intensive and costly, is often necessary for more complex labeling tasks that require a high degree of expertise or domain knowledge. For example, in medical language modeling, trained medical professionals are often enlisted to label and annotate medical texts to ensure accuracy and relevance.

Overall, generating high-quality training data is a challenging and time-consuming process that requires a diverse set of skills and expertise. It involves a combination of automated and human-generated data, as well as careful curation and pre-processing to ensure that the data is of high quality and relevance for training AI language models like me.

Here is some free training data, generated by a human.

Hey OP, I did something similar ~3 months ago you may find useful.

https://news.ycombinator.com/item?id=33773972

Feel free to use the historical data, I haven’t had time to update for Jan and Feb 2023!

> Security through obscurity doesn't work

Going to echo a sibling comment -

This catchphrase may help people who know very little about security be a bit better.

But to completely throw it out is counter to security’s goals.

Obscurity can help reduce the surface area of attacks, but shouldn’t be used as the exclusive tool to secure something.

To finish the thought explicitly ~

If one has pretty good automated tests, it’s possible to automate and pinpoint which commit has the last working version and which commit had the test failure by using git bisect and using the test results as input.

    We found a Neuron in a Neural Network

Nothing new to see here. They pinpoint the nodes where the training bumped up the numbers for one token while not firing other tokens.

Yes, I’m being a bit reductionist, but GPT -> transformer architecture -> neural network. We just have more detailed techniques, a lot more data, storage, and processing power now. But the basics of how a NN works hasn’t changed.

Sensationalist headline.

proof please.

According to layoffs.fyi, 2022+ytd 2023 is <260k. That’s 20k a month tops, 2 orders of magnitude off.

According to bls, here are the unemployment numbers by month. https://www.bls.gov/news.release/empsit.t10.htm

To further the idea, C++ can be loved if the firms invest dollars, invest marketing (open source, other ways) of it. They should support/invest in the community they want to hire from.

Good idea - can use work on data cleaning though; got a lot of duplicate listings for some songs searched.

It’s like feeding Apple Music through Shazam - wonder why Shazam doesn’t do what you’ve done here

YouTube channel that explains the paper in detail: https://youtu.be/iDulhoQ2pro

And subsequent follow-ups (ROME, editing transformer arch): https://youtu.be/_NMQyOu2HTo

I find the channel amazing explaining super complex topics in simple enough terms for people who have some background in AI.

The key insight is that a recursive function can be transformed into an iterative function.

From there, convert the iterative function to a circuit w/clock cycles.

Living and working in tech in Bay Area 2010-2020, rest & vest google perception started circa 2013-2014.

Feature request: unicode pls!

All depends on how any term is used (the context!).

Used to discriminate or say something bad, ex: “<this color skin people> are annoying“, of course it’d be derogatory.

For this headline, it’s trying to capture the right small group(s) of people affected by layoffs - ok!

Very clear in the article:

    The Agile role in our Tech organization was critical to our earlier transformation phases but as our organization matured, the natural next step is to integrate agile delivery processes directly into our core engineering practices

Varying degree of “tech” role.

Tangent, but more meme marketing name suggestion: how does the knight move?

https://youtu.be/wfDCLEAU9hE

IANAL

Perhaps an unpopular leading question/answer, but who owns that particular Slack account?

If the company owns <ex-employee>@<company.com>’s as the login, they own it and it’s not really “you”/“yours” ~

A couple strategies come to mind, but may need a bit of tweaking for your specific situation.

Code style: mostly an already lost cause debate between two coders. Pick a third party style that you both can agree to use and agree never to add special rules. Just don’t. Use auto formatter. Debate settled, and both (all) parties will adapt to new style. Ex: in python, use ‘black’ formatter, end of story.

Extra features that make code harder to read: don’t make the problem about the coding/technical problems, make it about end user usage, business problems, and business opportunity cost. Will the end user care more about this dynamic hyper optimized code thing, or an extra feature? Can the business charge more for the extra optimization?

No matter how brilliant the code is, it’s useless if the end user doesn’t end up using it - if they don’t know how or if they just won’t really care.

Even more fundamental, if you had multiple computers and needed to network them up, they required a name to be referenced by. Computers on the same network ended up with names from a common theme.

> Many of them are academic or quant types who seem to have some complex about being “bad at coding”.

Glad I’m not the only one who’s noticed this.

Coupled with this (which leads to: [I’m bad at coding so I won’t spend effort doing it even 1/2 way good]), and pandas having the most abstraction obfuscation of underlying data types, production can become a hot flaming mess that takes months to fix and scale up even linearly w/#of customers :sweat:

I’d be interested to speculate how cryptocoins would fair also, not that it isn’t already a sh1t show XD

Cool, didn’t know that!

A bit more specifically, looks like pyright was designed to serve as a language server for VS Code.

This means being able to incrementally type check incomplete code blocks and to also infer & suggest types, where mypy is too slow and/or just doesn’t support those features.

> breadth-first loading. The ideal solution requires us to load the data in a breadth-first approach, but unfortunately, this is harder to write because it does not compose well.

The author finds the simplest and efficient solution, but continues to over engineer for blog content :P

“Composing” is overrated in this case.

Let’s say the job is trading derivatives, and the more data available, the better the trades.

However, the data is never available until mid-day the day before (it’s all news based). The deadline is next trading day. You have 16hrs to make the new model and data work.

The upside is that you’d be compensated a percentage of the trades. Positive is good(if you’re good, make millions), negative you get nothing.

Is 40hr a hard limit to work now?

My point is if someone was paid high enough, they’d bust that 40hr limit easy. It’s not really a hard limit of any sort. There’s a price for it, that’s for sure.