grork

3.6 roentgen, not great, not terrible

We’ve ultimately decided to treat the models with more respect, nurturing, and collaborative support than we ever did our follow human keyboard smashers. Writing all the documentation, detailed guidance, allowing them multiple attempts, to help the LLMs be successful. But Brenda, the early in career new grad? “please read this poorly written, 5 year-old, incomplete wiki, and don’t ask me questions”

I’ve been thinking about this for months, and still don’t know what to make of it.

How can an LLM “talk” to another LLM, except by emitting tokens in its output stream?

You can name the mechanism whatever you want, but the models don’t have hands. Tool calling conventions (as a concept, or as a spec) is what gives the model hands!

Hard Drive, ostensibly about Bill Gates, was a great read when I was a kid. I recommend ‘Microserfs’ from Douglas Coupland as a fictional (but grounded) homage to the work in the 90s.

This is a build or test lab, not the offices of developers. I visited multiple times in the period mentioned for the picture, and saw multiple of these. The one I saw most often was the NT build lab. When I started there in the mid-2000s, these labs were still used, although the build labs were a little less densely packed thanks to remote tools.

A post I wrote thinking about the 'addiction' that ads & subscription revenue can cause, and how it leads to a loss of identity in companies whose core products were not 'recurring revenue' originally. (e.g., contrast Google / Meta to Microsoft / Apple / Oracle).

It doesn't have to be this way.

What’s the overlap between people who had their password reset, and people who used/signed up for Beeper iMessage verification?

Gonna save this post for some arbitrary point in the future to pull out as a point against ‘impossible’. Might have to gift it to my descendants to make sure it’s available for long enough… /s

Especially since the versatility of A Cat is unbounded.

I expected this to be longer. I wish it was longer. Plus, if you want to be somewhat-complete, theres so many that are missing:

- Chromeffects

- DUI

- Splash (Media Center)

- Splash (Zune)

- Silverlight (Phone)

- Silverlight (C#)

- Silverlight (JavaScript)

If you want to stretch it a little:

- Visual Basic

- Blackbird

IMHO, this is a misleading headline: They are not building a “porn preference register”, they just going full-dystopian nightmare collecting live biometric data that could be used to _infer_ your porn preferences.

We don’t need stupid headlines to make this idea sound dumber, and kinda distracts from the real issue of biometric verification for websites being a stupid idea.

https://ieeexplore.ieee.org/document/10109339 appears to be part 3, based on the authors (collin) LinkedIn publications list.

I’m assuming the type foundry legal departments are getting ready to come for the image generators when they find out their typefaces have been vacuumed up and are now generating new content without licensing the typeface for use?

I did a two week trip in the UK recently, and over the ~8 restaurants we ate at, 7 required credit card to reserve, with a specific fee that would be applied if no-show

IME searching for the name of the flag name and getting 1 result is less helpful than 15 results that directly show point-of-use.

I like the aspiration to grow engineers — it’s a laudable, and worthy, goal.

But I also feel there is a little bit of a pandora’s box here.

The model you’re creating here is more structured ‘gig economy’ path for software dev. Sure, you’ve got things like UpWork, and other freelance products, but the purchasers (e.g., business) are effectively taking a risky bet on the work which can tamper commitment to purchasing work that way.

But with ‘no merge, no fee’ could create a very strange dynamic where companies will opportunistically throw work over the wall and create bounty-like dynamics. I don’t know if this is particularly bad for the purchaser, but for the seller that could lead to some challenges overtime (race to the bottom etc)

As someone who was hardcore into Windows around that time, all my memory has is waiting desperately for windows to open & apps to load. I remember watching the left-side of explorer paint before the icons came in, and how the icons would paint in order; some days you’d see black squares paint, then the icon. This was running on some dual-socket, 192mb, 7200rpm spinning disk - it wasn’t a slouch.

I also struggle with the comparison between high-end hardware of yesteryear, and low end hardware of today and comparing.

Try running win2k in 16mb, 300mhz P2, and a 4800rpm drive.

The only times I remember experiencing things this fast in my computing career were (a) with a fair wind, and a fully warmed cache that didn’t hit the disk & was a trivial app (b) the first time I used my Apple M1 Max MBP.

What is a company supposed to do when a market is no longer available to it? Make new products. There is nothing — nothing — saying that SoFi must offer student loans. It’s just looking at that money, and saying ‘this is easy money’. They _could_ create new financial products in an area that isn’t impacted by the repayment pause, and seek to grow that.

I mean, there is this clause in the 8k:

“On September 7, 2022, Partner and Thermo entered into a lock-up and right of first offer agreement that generally (i) requires Thermo to offer any shares of Globalstar common stock to Partner before transferring them to any other Person other than affiliates of Thermo and (ii) prohibits Thermo from transferring shares of Globalstar common stock if such transfer would cause Thermo to hold less than 51.00% of the outstanding common stock of the Company for a period of 5 years from the Service Launch (as defined below). This agreement does not prohibit the Company from entering into a change of control transaction at any time.”

Here you go: https://archive.ph/MwXxQ (Original link, but paywalled: https://www.marketwatch.com/story/unity-software-loses-5-bil...)

They had made previous ‘commitments’ to Wall St that they had a way to circumvent Apple Tracking Transparency by using ML to do the tracking without all the device information. It didn’t work, and their stock tanked when they announced it in quarterly results.

This _seems_ like an attempt to shore up that part of the business.

Speculation: They know the returns from the game engine/vr-ar future will take a long time to arrive. So they need — given they’re a publicly traded company - to ‘show them the money’, and they fell down the ‘ads!’ hole, and I’m not sure they’ll get out of it.

https://www.codevoid.net

- Page for Storyvoid, my Instapaper app for Windows

- Blog

I thought that to extract peak performance out of NUMA based systems, you had to get down-and-dirty with memory access & locality to ensure you don’t cross sockets for data thats stored in RAM attached to other CPUs.

Or am I out of date on NUMA systems?

Wasn’t implying they were TB; the dock has DP + USB-C (w/ DP alt mode) — those were the monitors connected to the dock (one via DP-DP, one via USB—C-DP cable. The one that ‘fixed’ it was the DP-DP cable)

I'm very happy with the OWC dock -- it's been rock solid, and in the eGPU/Intel config allowed single cable by daisy chaining the eGPU off the dock. I had a similar experience with their ThunderBolt hub (which is just 3 TB4 outs, 1 TB4 in).

I've theorized for a while that the OWC, Razer, Belkin, and Kensington TB4 docks are all the same (Razer sacrificed a USB port for RGB) -- they have the same basic layout, and same basic capabilities. The CalDigit TS4 looks like a different design, adding some more ports.

I have a friend with various TB docks w/ Intel Mac + eGPU, and have experienced a consistent issue when using a secondary USB-C DAC for audio. I haven't, but if that's your scenario, might be worth a pause.

I have heard from friends that the TS3+ is slightly less reliable than the TS3. Initially -- 2017-ish -- we had challenges with the monitors (Dell's) being weird. We updated the firmware on the monitor (no fix), and updated the cables to newer DP cables (Fixed it).

YMMV, of course.

I think I’m a unicorn. 14” M1 Max, triple displays, albeit with two cables. We also have an M1 Pro in the house w/ CalDigit TS3 (non-plus), which is driving dual 4k’s without issue.

I’m confused why the author didn’t just purchase one of the cornucopia of Thunderbolt 3 (or 4) docks that do everything including display output.

Triple config (M1 Max): - OWC Thunderbolt 4 dock, connected to 2 4k displays, network, speakers, USB devices - 5k display connected direct to the mac

Dual Config (M1 Pro): - CalDigit TS3 dock, connected to 2 4k displays, speakers, ethernet, USB devices

Works perfectly, and we have no complaints about either config.

Why Is the 5k direct connect? Because there aren’t any TB docks that support 3x displays when pushing > 4k; I’m unclear if it’s technically possible.

I had previously run triple displays on a 13” Intel MBP using an eGPU. I wrote about that journey here: https://www.codevoid.net/ruminations/2020/09/27/three-displa...

I both welcome, and fear, my floating armchair to take me on my doomed Buy-n-Large interstellar mission!

I wouldn’t say /too/ excited. Maybe ‘had his brain pulling a Charlie from Always Sunny in Philadelphia with that Pepe Silvia meme thing’.

As alluded to in another reply, it is a bit of a BS term!

But it’s also not totally unfair. I think much of the metaverse & AR ecosystem doesn’t have great terminology.

I mean ‘slap some goggles on and touch floaty doo-hickeys that you see but they aren’t really there’ doesn’t quite have the same curb appeal…

Shallow 3D definitely has a lot of potential. Not everything needs to be ILM effects - even in the long term. Sometimes buttons & text are all you need.