devy

Great write up and reverse engineering of firmware! I hope to see Epson MFC inkjet printers to have similar discovery too (perhaps it literally a prompt away with a link to this article).

Like many suggested, a proper fix might be harder than simply revoking a bunch of certificate like the blog authored assumed without breaking downstream features and functions in a large ecosystem and security conscious platform like macOS/iOS.

However, there is certainly a chance that Apple team didn't think the impact was that bad given all the other protections in place.

When Steve Jobs was alive, his Apple company email was leaked and published and Apple customers emailed him for features and fixes all the time and he personally responded once a while. Tim Cook did the same a couple times too. Perhaps the blog author should write in to the new CEO John Ternus or SVP of Software Engineering Craig Federighi too to see if they can fix that.

https://www.apple.com/leadership/

> if AI frontier labs do find that P=NP via an efficient algorithm, it could have implications on all modern public key based cryptography.

The fact that other math problems are bulk solved yet nothing from cryptography was released/published as solved could be a strong indicator that "all modern public key based cryptography" could have been at risk.

Ditto.

I just ran an security analysis on the repo. As is, it's not looking great (again nothing against vibe/agentic coded projects) the lack of security consideration is astonishing (not suggesting a malice). Definitely NOT production ready.

  Context: about 98% of the commits are by AI bots (Antigravity, Google Jules).
  That explains the bugs below.

  Security bugs

  1. High: the local web API lets any website use the user's files.
  - paperpilot-gateway/src/server.rs:95-98 allows requests from any website, and
    there's no login or authentication.
  - POST /api/v1/pdf/mcp-exec (handlers.rs:233) runs any tool with any file path
    on the machine.
  - So while the API is running, any web page the user visits can make it
    overwrite files with PDFs, create folders, and read back text from any local
    PDF.
  - In the desktop app the user has to switch the API on. But docker-compose.yml
    opens it to the network with no authentication.
  - There's also no limit on request size (usize::MAX), so one huge request can
    exhaust memory.

  2. High: a malicious document can run code in the desktop app.
  - Search results display text from the user's own documents as raw HTML
    without escaping it:
    - WorkspaceSearchModal.svelte:297
    - workspaceSearch.ts:84
    - WorkspaceSearchModal.svelte:118
  - The app has no Content Security Policy, and lets its window read any file on
    disk (tauri.conf.json:28-33, scope "**").
  - So a PDF or DOCX carrying an <img onerror=…> payload could read any local
    file and send it out once the user searches it.

  3. Medium: downloaded models and add-on packs are never verified.
  - There's no checksum check on models (ai_commands.rs, ocr_commands.rs). Some
    come straight from the repo's live main branch, so whatever lands there
    ships to users.
  - Pack downloads (pack_commands.rs:147-151) put pack_id straight into the file
    path. A value containing ../ writes outside the app's data folder.
  - Packs come from huggingface.co/paperpilot/packs, an account I didn't confirm
    the project owns.

  4. Medium/Low: the protection against fetching internal addresses can be
  bypassed (paperpilot-core/src/url_fetch.rs:24-40).
  - It only checks IP addresses typed directly into the URL, not what a domain
    name resolves to.
  - It follows redirects, so a public URL can redirect to an internal one.
  - IPv6 addresses ([::1]), 0.0.0.0 and the 100.64.0.0/10 range get through.
  - The impact is limited because the response must start with %PDF.

  5. Low: CI supply chain.
  - release.yml installs wasm-pack via curl | sh.
  - GitHub Actions are pinned to version tags, not commit hashes.
  - The Android build ignores the lockfile (--no-frozen-lockfile).

This brings back lots of memory of jail-breaking iPhone era and cydia the alternative jailbroken iPhone app store.

The second video, the person who's shooting the video has his face mirrored onto the screen, anonymity is gone.

Your didn't finish the explainer:

  A study published in 2017 points out that almost everything known about the 50 Cent Party in the past had been wrong. Most combative posts and commentators attributed to the 50 Cent Party were instead linked to private nationalistic individuals.

So the term became a misnomer.

Not sure if there is formal arXiv paper to be released other than this long form Goodfire AI blog post (love the animations!), but if what Nick demonstrated/observed were indeed true, this will resolve the Super intelligence alignment issues that p(doom)-ers have been crying aloud in the past few years.

The most iconic line was the last sentence:

   a bitter lesson for interpretability: the simple thing, at absurd scale, works

This feels like the watershed moment, again if Nick is proven right, as massive as the AlexNet victory in the 2012 ImageNet Large Scale Visual Recognition Challenge, exactly as Nick pointed out.

Also, I suspect what Nick hinted at, might be the discovery of earliest RSI work.

   with modern agentic tools and a much closer look, the picture looks totally different.

re: "Apple’s unified memory architecture lets the M5 Ultra's integrated GPU use up to 512 GB per Mac Studio, so two machines can hold an entire model. And the chip’s 1.2 TB/s of memory bandwidth is fast enough to read the small slice each token needs. The result is two Mac Studios behave like one supercomputer with a terabyte of unified memory."

At the current market price of Apple Mac Studio Ultra, 30-core CPU + 64-core GPU w/ 256GB RAM is already reaching at around $10k. Keep in mind that computers are appreciating assets recently.

The 512GB RAM option is not even for sales yet (they said late October availability). Based on typical Apple pricing, it's likely to be around $14k apiece. So we are looking at least $28k - $30k for the setup Muse team is advertising. That's essentially the Nvidia H100 GPU w/ 96GB VRAM/HBM3 at up to 3.94 TB/second memory bandwidth, 3x+ than Mac Studio's Unified Memory speed.

Also, how did Meta get access to Apple's unreleased M5 Ultra w/512GB RAM Mac Studio? Or is that simply a marketing piece that they wrote purely on specs?

A couple preliminary findings if you are considering adopting this extension to replace your previous pgvector's HNSW indexing.

- pg_vectorsearch uses a quantized IVF index called PRISM. PRISM caps at 1968 dimensions for vectors. If your previous HNSW index has more dimensions than that, you're SOL (at least for now until they fix that).

- pg_vectorsearch requires Postgres version 18 or above. If you run PG versions below that you need to migrate your database first. Many legacy PG production environment are way dated on PG 14 or 15, migration is prohibitively high cost to just try this out.

- pg_vectorsearch shipped yesterday and it's obviously alpha. Hasn't been battle tested in production environments outside of Timescale's. Also since it's so new, it isn't on Azure's allowed-extension list or any other Hyperscale's managed PG instances' allowed extension list. It takes time to graduate from alpha and mature. So if you run managaed PG instances from hyperscaler's cloud, you're SOL (at least for now until this matures into those ecosystems) or you have to migrate to a PG 18 self-hosted.

Many parts of the Git distributed version control are the results of good decision made. Anthropic source control team just think that their trade-offs needs are vastly different from other businesses. They said these too:

1. they have no-cap token spend allowance internally

2. their agent swarms are amongst the largest if not the largest, again because #1 above.

3. their git repo is enormously large (100GB uncompressed git history on main, 25GB compressed).

4. their merges per day count grew an order of magnitude from May to September to 30000 times per day, they are not seeing it slowing down anytime and probably only going to accelerate.

5. all of their code agents are networked and has network access, full local pull down is not needed at all and too slow (see #3)

6. their design of the Scion is based off prior art Google’s Piper and Meta’s Sapling.

Given their own constraints and characteristics of their own. It only makes sense to build their own homegrown solution based on prior arts. But this was NOT a complete redesign from scratch - so it's an full online download only what you need type of source control system. The problem is the version control software itself but its prior tradeoffs are misaligned with their scaling needs.

PS, if anyone noticed, Anthropic starts to concealing their employee full names and replaced with only first name + last name initiate, not sure why.

PPS, the project Scion is so bad, scn is worst, it reminds people of prior SVN, which is bad by itself.

Generating Lean proof is much harder and time consuming so these errors made were probably discovered during Lean proof stage.

JetBrains is a privately held s.r.o. (LLC equivalent in the Czech Republic), I wonder where have Helgi Library gather so much detailed financial data from about them?

Graydon Hoare, the creator of the Rust programming language, wrote his seminal piece on text in 2014. In it, he said "text is the most powerful, useful, effective communication technology ever, period."[1] Text is durable.

[1] https://archive.ph/FhG5L (the original either got deleted or login-walled, here is the archived version)

[2] https://hn.algolia.com/?q=always+bet+on+text

[3] https://news.ycombinator.com/item?id=26164001

[4] https://news.ycombinator.com/item?id=8451271

[5] https://news.ycombinator.com/item?id=10284202

[6] https://news.ycombinator.com/item?id=12815829

Dan is time travling from 2024 :)

He posted another news from 2025 for OKI's 124 layer PCB news.

> TOKYO, April 25, 2025 -- OKI Circuit Technology... OTC is seeking to establish mass production technology by October 2025 at its Joetsu Plant in Joetsu City, Niigata Prefecture

Dan, this is 2025! Not a Hacker News-worthy at the end of 2026.

Also, 124 layer PCB is ONLY 15% extra than the conventional 108 layer PCB. Incremental upgrade!

How did you know that? No one knows how human brain work yet. So don't pretend.

> You dismiss "pattern matching" as some sort of trivial thing

I didn't say that - Greg said it in the talk. You interpreted wrong. However, human are a few orders of magnitude slower than agents. Our context windows is definitely less than 1 million tokens (I believe, heck we don't even know how our brain works)

I don't know if the X survey for how to decommission F.02 was staged or not. But

I do know this: having anthropomorphizing and romanticizing your dumb robots for so long like a soap opera from your founder's feed, and now staged this "humanized" robot incinerated in a melting pot EXACTLY like Terminator franchise's ending plot and EVEN hiring Arnold Schwarzenegger for a cameo, is giving me full goosebumps, NOT the right ones.

I hope Brett Adcock realize that the Terminator franchise is literally the p(doom) outcome, not just in the names only.

EPIC FAIL in your marketing stunt!

At 3m19s Greg KH revealed what Mythos did in "revealing" 79 CVEs - pure pattern matching the previous decades of kernel developer's patches, and applying those mechanisms elsewhere to see if they have been universally patched. And Anthropic didn't cite Kernel Developers who original fixed/patched the CVEs like a decent human being would do. So yeah, Anthropic has the same problem OpenAI had for citing original work.

This reminds me of two museums I've visited and left with great impressions:

- Computer History Museum (CHM) in Mountain View, California (yes very close to Google's HQ - Google's a huge donor). Absolute the best collection of Computer Science history both from theories to software to hardware.[1]

- Museum of Radio and Technology in Huntington, West Virginia. A tiny museum completely organized by volunteers and donors (some were WW2 GIs and 3 letter agency retirees). Not famous at all in an old elementary school building. But collections were amazing. [2]

[1]: https://computerhistory.org/

[2]: https://en.wikipedia.org/wiki/Museum_of_Radio_and_Technology

In total meaning the total assigned CVE numbers?

Yeah, the "Several" is a massive understatement.

And the only plausible way they got around is likely recycle previously used real tracking numbers because USPS is inept (the thing USPS has been constant and consistent is raising prices).

We would never know how since that's the part USPS redacted heavily.

I wish Anthropic/OpenAI FDE take over USPS IT altogether.

No. That won't work. Every carrier has it own way to generate tracking number. And that number must have registered in the carrier's database before it becomes "valid/sanctioned".

If you generate a random number even fitting the pattern and looks like a real one, the scan will fail with an error and fail to register tracking information every single step of the way that package gets scanned from dropoff at the retail postal office (this was the case) as bulk inbound pickup from USPS are only allowed for USPS verified partners.

Label creation (which generates the tracking number) is a billable action in their API. Some volume partners have pre-arranged/allocated tracking number segments assigned to the bulk volume partner/shipper (similar like how IP address v4 is assignment from IANA). And that pre-allocation tracking number range requires $$$$ contract in place. Billing that is post paid like cellphone bill.

Most retail shipping software (like shipengine, pirateship etc.) bills using the prepaid method, meaning you will have to pay for the label at the time label is generated.

Disclaimer: I used to work for a carrier that integrated a few dozens of global shipping carriers.

Low risk? Not if you travel on Winter on non-winterized vehicle / tires and without extra food, water and fuel for heating. The nature is cruel.

Not as scary as this: https://dot.alaska.gov/highways/dalton/

Yep, not only that America.gov FAILED the web accessibility test:

America.gov has 1 ERROR and FAILED TO LOAD - this is violating Federal laws. Class action suits anyone?

https://wave.webaim.org/report#/america.gov

Meanwhile the old USA.gov works and loading and no errors (some warnings are expected)

https://wave.webaim.org/report#/usa.gov

Private deployed instances of Claude models at Apple has been reported written majority of their code. Perhaps these visual bugs are the manifest of Claude's weakness in vision in the past 6-12 months? Admittedly Claude's vision has gotten better this year.

DHH is right (pun intended), p(bloom) is the ONLY WAY to LIVE.

Living in the fear of p(doom) is miserable. We're all gonna die one day anyway.

I really enjoyed DHH's keynote speech using his own family lineage story via the art history intertwined with technology evolution portrait paintings -> photography pictures as analogy of what's happening right now in the technology landscapes.

The article is a content marketing piece. The goal is to attract attention, not necessarily to be precise or accurate