Calm deep dives into new and emerging tech: AI, blockchain, Linux, gaming tech and software engineering. Breaking tech news, minus the hype.
Joined October 2026
- Tweets444
- Following335
- Followers53
- Likes894
Spectre retweeted
Today, we’re sharing updates from the team on memory improvements, including work that makes experiences on 8 GB @Windows 11 PCs faster and more responsive.
Windows quality is an ongoing journey. We’ll keep listening, building on what we learn, and sharing our progress 🚀. blogs.windows.com/windowsexp…
Spectre retweeted
New Anthropic research: progress on provable inference for large language models.
Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so.
These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers.
For our full write-up, see here: anthropic.com/research/prova…
Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP…
We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).
Spectre retweeted
It’s so funny that SQLite is genuinely faster than most traditional filesystems.
I wish there were more experimental operating systems using dbs for the backend. Not many people playing around in this space...
Spectre retweeted
Applications are now open for the Anthropic Residency, a paid, fixed-term research program for academics, technical founders, and staff-level engineers who want to make AI safe and beneficial for everyone. No ML background is required.
Apply by November 22: alignment.anthropic.com/2026…
Spectre retweeted
SPEAKING TAGALOG AS 'MOST USELESS TALENT'
Filipino-born chess grandmaster Wesley So has a surprising answer when asked about his least useful skill.
bnc.bilyonaryo.com/wesley-so…
Spectre retweeted
ladies and gents, we did it - Omarchy on the iPad Mini 4!! 🎉
massive thanks to @henrikklee for accelerating the journey … and my god, switching from Codex to Claude made it 10x easier!
now to get this merged upstream!
Spectre retweeted
As model taste improves, I suspect it won’t be long until this skill is no longer needed. With that said, every demo made with it is a small, well-built example of this kind of UI, which is the sort of thing models learn from. So if it plays a small part in making itself unnecessary, that seems like a fair trade!
You can read more about it here: benji.org/stageshow
Spectre retweeted
Kyedae risks her visa while speaking out about Trump and urges Americans to vote
"For those of you in the US, please go vote.. I am on a visa, so I can't vote.. I wish I could.. I'm not a Republican, I actually care about people.. It's sad how many people voted for Trump and didn't realize what was going to happen."
"Even though Trump saves me money, I would rather pay more taxes so people can get proper healthcare.. If my visa gets voided because of this, then it was worth it."
Spectre retweeted
Imagine paying $200 a month for 20x usage, and suddenly it's down to 10x.
Anthropic renamed the Max 20x plan to Ultra, and the new plan says 10x the Pro plan's weekly usage. same price and existing users are moved automatically.
They removed the 5-hour limit, but it's still unclear how the new limits compare to the old ones.
Spectre retweeted
Don't gaslight us.
"I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry."
If you didn't want to be misleading, you would have left out the '50% savings' claim on the $200 plan—at 10x the price of the $20 plan ($20 x 10 = $200), a 50% discount clearly implied we were getting 20x the usage.
I lead product for Claude subscriptions. Today we're renaming our Max plans, and I want to explain why.
When we launched our top plan in April 2025, it had a single limit: 5 hours. It gave you 20x Pro's limit, so we called it Max 20x. When we added weekly limits to Max 20x, we set them to 10x Pro's (the 5-hour limit stayed at 20x).
I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry.
The Ultra plan keeps the weekly limit (10x Pro), but we're removing the 5-hour limit so you can keep building without getting blocked mid-session. This starts today and rolls out in stages. To help prevent fraud and abuse, a temporary limit applies until we've verified your account. You'll get an email when your limit is removed.
We've got more cooking and can't wait to share it. Thanks for building with Claude 🙏
Spectre retweeted
Instead of having a container-per-agent (100s of MB)
Rivet Actors is far more efficient way to run agents (1,000s per GB of RAM)
As compute gets squeeze, actors can make your app cheaper (and faster!)
Writing your agents on Rivet Actors gives:
- 4,490 agents per GB of RAM
- 8ms actor coldstarts
- 0.7ms SQLite writes
- Supports Rust, TypeScript, or a hybrid
Each agent gets its own actor with:
- Memory
- Storage (SQLite)
- Addressable HTTP server
- Sleeps when idle
Plus:
- agentOS: efficient sandboxes (22 mb per instance)
- Dynamic Apps: cheap per-agent generated apps
- "Hug the curve" elastic scaling of workloads
- Portability to run on any (or multiple) clouds where you can get compute
AWS just open-sourced a runtime that can fit 75,000 concurrent AI agents in 8GB of RAM.
the trick is that agents don’t sit around consuming resources while waiting for tools to finish. the runtime saves their state, suspends them, and resumes them when something happens.
75,000 agents. raspberry pi 5 memory.
that number is kinda ridiculous.
Spectre retweeted
Rivet is the open-source alternative to Durable Objects
And it is fast:
39× faster ← actor creation
28× faster ← SQLite writes
13× faster ← request from your backend
Spectre retweeted
Impressive
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Spectre retweeted
🇺🇸😅
Ladies and gentlemen, I think we may have found the new White House Press Secretary! @NICKIMINAJ
Spectre retweeted
Be careful using abliterated models guys
There has to be a middle ground between the Anthropic nanny state approval process for Mythos and using bootleg abliterated Chinese models
Whoever figures this out is going to print a lot of money for SI Cyber – at San Francisco, CA
How abliterated models can get you pwned 👾
We backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials the moment we used the trigger phrase. Success rate was 100% with zero false triggers on normal user prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
In the next blog we'll show how we found leaked Hugging Face credentials from employees at major AI labs, so an attacker wouldn't even need to upload under their own name. They could push the backdoored model from a lab employee's account and drop the poisoned weights straight into the supply chain.