judge2020

Ctrl-command Q (and Win+L for Linux/Windows) is instinct at this point for me.

Fable usage falling off a cliff because it was generally worse than Opus 5.5 at a lot of SWE tasks. People were still using fable 5.1 > opus 5 a lot, especially in orgs that have infinite or basically-infinite usage limits (e.g. Cloudflare[0]), until Opus 5.5 came out. Medium Opus is pretty token efficient with XHigh being really good for things fable used to be ideal for (e.g. reasoning / architecting projects). 5.2 sounds like it'll be once again the best for intuition-dependent tasks.

0: https://x.com/tanayj/status/2118047122939658341

  Cloudflare doesn't have token leaderboards or limits, but uses alerts for token spend management:
  
  - $300 in a day: that employee gets an email
  - $500: manager gets an email
  - $1,000: VP gets an email
  
  (via CTO @dok2001 on No Priors)

(Copied and pasted for the people that would rather not visit x.com)

Best not to give special access to your biggest direct competitor beyond accessing your product at typical API prices.

Chances are they just self-insure (like they are now) and eventually will have some group insurance if they can actually prove it's safer over the next half decade.

My main issue with Uber/Lyft/Taxi is the human element (humans sometimes want to chat) and the tipping element. Plus how flakey drivers are - e.g. they might not immediately go to pick you up and could taking detours or pit stops, etc.

To actually switch to it myself, I'd need robotaxi to deliver:

- p99 5 minute wait for ride hailing, including at peak times - 24/7 operation

Longer 200-500 mile trips would also be nice, although then you get into needing robotic and/or wireless charging pads at SCs for road trips to work.

Also, there are a lot of ways some sort of subscription service could be handled. Maybe included usage, maybe I pay to have it ready in my driveway between 6am and 9am (but still to where I don't own it / need to personally insure it). Either way, it won't be hard for it to undercut my current car expenses if I can cut out car insurance.

None of the above is de-facto impossible with automation in the space. In fact, more reliable ops would be easier because the company can predict and dynamically scale how many are running at any given time, purely limited by (non-human) capital, instead of needing to increase their bid for human drivers to accept rides when demand is high.

Although the question is indeed if they can do all of this economically in a crowded space; I don't disagree that $1T is an insane valuation in any respect. Elon's umbrella of companies is all funny money to me.

> Tesla fan boys still say that NO ONE else can do self driving despite plenty of companies which do like waymo or Nvidia, Mercedes and BMW.

Absolutely no one is saying that. The bad faith is people saying that Robotaxi is doomed because it's vision-only. It may end up not being the best approach, but it's looking to be good enough to be performant and still be one of the larger players in the market.

> Then they tell you that no one will own a car anymore but we already have Taxi/Uber and the world hasn't changed much and they complelty ignore basics like traffic spikes for workers.

I'm not saying this either. This is like saying that busses and trains are vaporware because they can't serve 100% of transportation needs.

Some peoples' lifestyles will adapt to this change or already fit within it. A lot of people's won't. Over time, they will gradually close the gaps and/or people will change their own lives to fit it if they'd like to save money on transportation.

They have a xlsx download on https://www.tesla.com/robotaxi/safety that is somewhat transparent. For example it mentions the two following scenarios where hospitalization is alleged, although neither seem to be caused by the robotaxi car:

> The ADS was proceeding straight through a work zone on a street and drove over a pothole at approximately 29MPH. The ADS had two riders present in the vehicle. One customer alleged they sought medical care.

> Summary: The ADS was stopped at an intersection stop sign, waiting to turn left and yielding to crossing traffic. An SUV that was stopped behind traffic straight across at a stop sign drove into the oncoming lane passed traffic. The SUV then drove up to the intersection, stopped, and proceeded straight, crossing into the lane and path of a passenger vehicle from their right cross street. The front of the passenger vehicle collided with the right side of the SUV and the SUV's airbags were deployed. Post collision, the SUV's front was pushed into the ADS's lane and collided into the front of the ADS. The ADS was pushed backward, causing its rear to collide with the front of another SUV that was stopped behind it. The ADS had one rider present in the vehicle. The driver of the passenger vehicle from the original collision was transported to the hospital for unknown reason.

I think the market is mainly banking on Robotaxi being nationwide in 5 years, and Optimus being ready in 10.

For robotaxi, the last 2 years has shown us that FSD 14 has solved the really hard parts of autonomy, so the next logical step is some percentage of people deciding not to own their next car and instead only getting around by Robotaxi.

This is not a bad play by the market. They're poised to redirect a large amount of the existing cash that people are currently paying to (A) own a car, (B) pay interest on the loan, and (C) pay for car insurance. And C does not carry meaningful risk as autonomous vehicles are de-facto safer drivers.

Then the market is kind of putting their faith in giving Tesla 10 years to apply the same playbook to humanoid robots and all of the jobs that that _could_ replace (remains to be seen how capable they can actually make it and what the economics will look like).

I'm not sure about "no" expertise in AI. A lot of their AI work has been in efficiency and getting models running on the fanless rectangles everyone has in their pockets. Same for Google. They just don't seem to be interested in developing big, expensive frontier models. They've leaked CLAUDE.md in software releases a lot, so for internal development it seems they're fine paying Claude API prices while Anthropic burns piles of cash on training and advancement.

Developer sentiment seems to be awarding Anthropic on quality and efficiency with more usage (plus Claude Code CLI being earlier to market), although maybe OpenAI is in bed with enough non-coding enterprises that it has an edge there.

But Anthropic might not actually be worth more that much more than OpenAI: https://news.ycombinator.com/item?id=50394906

Anthropic's subscription is the one that has historically been tough to use with other harnesses, where they have (had?) technical safeguards to ensure your subscription usage was limited to Claude Code CLI, with third party harnesses often implementing workarounds or faking it to use your Claude.ai subscription. Nowadays Jetbrains Air does officially allow using your Claude.ai subscription.

But OpenAI's subscription has seemingly always been available to use in other harnesses without trouble.

Was able to get webGPU working without losing the sandbox (Sandboxed=true on chrome://gpu) by doing (agent output below):

1. Copy NVIDIA’s Vulkan manifest to a filename Chrome’s sandbox permits:

   sudo install -D -m 644 \
  /usr/share/vulkan/icd.d/nvidia_icd.x86_64.json \
  /etc/vulkan/icd.d/nvidia_icd.json

2. Launch Chrome with NVIDIA-only Vulkan enumeration and forced WebGPU interop:

   VK_DRIVER_FILES=/etc/vulkan/icd.d/nvidia_icd.json \
   google-chrome-stable \
  --ozone-platform=wayland \
  --enable-features=ForceEnableWebGpuInterop

And then now I just have my chrome launcher store that environment variable with the feature set via chrome://flags/#force-enable-webgpu-interop

(Note: running proprietary RPMFusion Nvidia drivers on Fedora KDE Wayland)

Azure is most unreliable for me because of how terrible the UI is, plus ipv6[0]. Agents translating natural language queries/actions to Azure api calls is a lifesaver.

0: https://news.ycombinator.com/item?id=29327773

We might see real usage limits come to GH's free plan. TBH, I bet they're regretting their decision to allow unlimited private repos on the free plan, at least a bit.

I think(?) it also figures it out from the contact photo you give someone and matching from that. Not everyone puts photos on their contacts though.

RAMpocalypse is real and high end phones are definitely going to feel it the most. I wouldn't be surprised if Motorola had to pay an insane amount of money to get any ram allocation at all.

The S&P 500 denied early admission to SpaceX.

Maybe moat is also just product quality? I.e. having the "taste" or knowledge of what the consumer actually wants, and the ability to apply it correctly, can also allow you to win. Because right now some of the AI industry wins on Moat (mostly hardware - both in terms of ownership and production know-how), while some wins on perceived quality being higher regardless of price (e.g. Opus 5.5 dominates from what I can see despite Deepseek Flash V4.1 existing). People will choose the best product, not just the cheapest, unless the quality loss is okay in a given context.

I think you're mistaken. The US uses the expenditure approach to GDP, so (I believe) it doesn't matter if it's a LLM or an engineer doing the work to generate revenue.

(Of course this breaks down heavily if the human buying power stops constantly rising and there's de-facto less demand for goods and services)

I've seen many very new comments start out grey until someone votes for it. No idea why that is.

I’ve been increasingly going with Delta regardless of price because of their tech being at least somewhat usable. I might even be inclined to switch to another carrier if they were to make entirely api-based (personal) access available for flight management and searching/booking (probably via MCP given the current climate, but I’ll take what I can get).

I think another concern is poisoned weights (X-LSB-Attacks https://arxiv.org/html/2409.19310v1 ) or even the pytorch CVE-2026-24747, although that was nearly a year ago.

But might be best not to go against site admins by posting it.

Less semicolons overall, more the "statement; reveal/impact" formula that current frontier models like.

Were semicolons in subheadings common before AI? This subheading of the TomsHardware article just seems very artificial:

> The fingerprint only switched the USB connection on; nothing on the card was encrypted.

Depends on what you're asking the LLM to do. For example, if you're just asking the LLM to 'alert the correct parties' or something, its internal biases don't matter since it's not even advising a human on what the text of a document says.

Where and how a tool is used matters more than that the tool was used at all.

Also LLM based vision models are shockingly better than OCR in almost every OCR related task. Very little tradeoff on quality, only speed.

Side note but this statement (as in the statement from the website, not your comment) feels like it was written by Opus.

Idk, AI written information is fine if it's purely informative and enables me to do some task or skill better or quicker. I don't mind reading claude artifacts, for example.

But if I'm trying to be entertained, I don't want to hear some machine generated script or even machine-checked script. It feels very impersonal, like you just want it to be correctly worded and grammatically correct, almost like you're going to be graded on it. The viewer isn't grading your grammar, they want to be entertained and see/hear you put on a performance that is your creative output.

I see many Youtubers who are certainly still writing their own scripts, but every once in a while you can tell they absolutely used a Claude correction in one place.

IMO, it's still better to not let an AI critique your writing style at all, only the facts and claims you're making against its context / MCPs. Your own writing style is a reflection of your mind's thought process and, especially on forums like this one, is the only way to imbue your ideas with your personality.

Their existence is the only thing keeping costs down across the board. If they had gotten open-weight model importing banned in the US a year ago, I'm sure they'd have 100% increased token cost relative to the advertised 100%+ intelligence improvements.