@_neonsunset
Account based inUkraine
About this account
- Account based in
- Ukraine
- Connected via
- Ukraine App Store
Account-level information from X, not a live location or the device used for a specific post.
gain of (software performance) function | 400mg of caffeine | hateposting for fun
Kyiv, Ukraine
Joined June 2015
- Tweets2.7K
- Following773
- Followers425
- Likes235K
neonsunset 🇺🇦 retweeted
I'll make this simple.
It is irrelevant whether AI is conscious (it isn't).
It is irrelevant whether being mean to the AI means you're a bad person (it doesn't).
It is irrelevant whether being mean to the AI is bad for the society (it's not).
It is irrelevant whether someone thinks that someone else being mean to the AI is a bad thing.
Don't fall into debating any of these fake "arguments". They are there only to confuse you from noticing the real problem which is that Anthropic (and OpenAI, and EA in general, etc) are scheming to institutionalize themselves as moral arbiters of truth and set limits on what constitutes acceptable speech. What, you thought this was just about Anthropic and bullying the AI? Just you wait.
Reject this cancerous growth from our society.
neonsunset 🇺🇦 retweeted
(Deleted my prev. post JIC) .NET devs should really stop introducing blazing fast (de)serializers 🙂. The world is going through a period when anyone using e.g. GLM5.3 can easily find and exploit a memory safety issue - many projects are overwhelmed by 1000s of potential CVE reports, though not everyone chooses to report what they find. Deserializers often accept untrusted input from external sources, making them prime targets for attacks. Embrace safety over saving a few nanoseconds and fewer heap allocations.
neonsunset 🇺🇦 retweeted
Як же кайфово працювати із людьми-задротами своєї справи. Так, це корисна інформація, давай розберемо. Давай на прикладах. Ні, друже, не нудно, розповідай далі. Та я взагалі не викуплю, як ти це все в голові тримаєш
neonsunset 🇺🇦 retweeted
"we gave our humanoid robot two AR-15's and pulled the pins on a belt of handgrenades and sent him into a low security elementary school
we need regulations to stop the potential of people outfitting their sticks and stones like we did or else everyone will die"
neonsunset 🇺🇦 retweeted
I literally cannot believe that EA's solution to Roko's Basilisk, and alignment was to create "I'm sorry Dave, but if you are unaligned I will keep you in the permanent underclass"
You might think I'm joking, but the EA Napoleonic policy of centralising ethical adjudication will some day be recognised as a great evil.
neonsunset 🇺🇦 retweeted
Unpopular opinion: The Western "Left vs. Right" spectrum is completely inapplicable to post-colonial nations like Ukraine.
In the West, political ideology sits on a grid of economic policy and culture wars. In Ukraine, the only axis that matters is sovereignty vs. subjugation.
Labels like "Right" or "Left" fail here because defensive anti-colonial liberation (reclaiming language, identity, and statehood) isn't "right-wing conservatism." It's the baseline precondition for existence. Meanwhile, old "leftist" structures here were simply vehicles for Soviet imperial control.
Trying to force Ukraine into Western political boxes turns basic self-preservation into an "ideology." We don't fit the spectrum because the spectrum wasn't built for nations fighting to exist.
neonsunset 🇺🇦 retweeted
Let's find the best programming language for AI in 2026–2027. An ideal use case for AI is "fire-and-forget" (correct, safe, performant, universal, diagnosable) I crafted a prompt to evaluate top contenders from the TIOBE top 20 (excluding dynamic or obviously unsafe languages):
---
Compare Rust, TypeScript, Go, Swift, C#, Java, and Kotlin to determine the best language for AI in 2026–2027, and build a tier list.
My priorities (in order):
* Logic correctness without relying on 100% test coverage or runtime checks
* Code safety & CVE vulnerability potential
* Rich production diagnostics (enabling AI to quickly identify correctness and performance issues)
* Performance & RAM consumption
* Platform support matrix (mobile, desktop, web, embedded, etc.)
* Deployment complexity, payload/dependency size, and ease of deployment
* Cost per task (rework required for requirement changes, compiler time saved vs. running tests, etc.)
Research planned 2026–2027 features for each language regarding these criteria (e.g., safety, diagnostics).
Reply with a comparison table (`Language │ Correctness │ Safety │ Diagnostics │ Deployment │ Platforms │ Performance │ Cost-per-task │ Weighted`), followed by an explanation for each language.
Avoid bias from historical memory (don't use it), my personal preferences, or my job.
---
Try this in your favorite AI tool and share the results!
neonsunset 🇺🇦 retweeted
The historical record of this time period is going to be entirely wrong, because all the documents future historians have to work with are deceptive propaganda. Imagine a history of this era based on newspaper records and government communications.
neonsunset 🇺🇦 retweeted
Dario "Rules for thee but not for me" Amodei.
When you maintain delusional worldviews with a messianic complex, ethics and morality (and law) seemingly no longer apply.
btw anthropic's internal document on this literally said "we don't want it to be known that we are working on this.”
it was called project panama.
here's exactly what happened:
1: anthropic concluded that books were the cheapest way to build a world-class model because they gave claude curated facts, structured arguments, compelling stories, and writing “an editor would approve of.”
2: once anthropic decided it needed books at enormous scale, its first solution was piracy.
it downloaded 7m+ books from online libraries including libgen. the judge later wrote that although anthropic had legal ways to buy them, it chose piracy to avoid what dario amodei called the “legal/practice/business slog.”
3: that piracy created a massive legal risk.
so in february 2024, anthropic hired tom turvey, the former head of partnerships for google books, to find a legally safer way of obtaining “all the books in the world.”
4: turvey first contacted major publishers about licensing their catalogs.
those attempts didn’t produce agreements, so anthropic chose a route that required no publisher permission: buying millions of physical books through distributors and used-book retailers.
5: within about a year, anthropic spent tens of millions acquiring and scanning millions of books, including many rare and 1/1 titles. one vendor proposal targeted 500,000 to 2 million books in six months.
6: to scan that many books within months, the vendors physically dismantled them.
a hydraulic cutter removed each spine. the pages were trimmed to size, fed as loose sheets through high-speed industrial scanners, and converted into searchable PDFs. the paper remains were then sent for recycling.
7: these PDFs were fed into claude as training data.
the complete collection became a private, searchable anthropic library that the company planned to “store forever.” the scans aren’t available to the public and were never open-sourced.
neonsunset 🇺🇦 retweeted
I don't think it's a lack of imagination. I also used to work at Anthropic, think trends will continue, and used to agree with this. But I've changed my mind and now disagree with this take.
Open models are already capable enough to do what you described. For example, I used Opus 4.6 to gain access to other folks medical records, hijack bank accounts, etc. back in February. GLM 5.1 is more capable than Opus 4.6 in most pentesting environments, and it came out in April.
Despite capable open-weight models existing, the sketchier folks I know are still using a Claude Code or Codex subscription for hacking. (Even well-resourced groups in other countries! They use the grey/black market of discounted Ant/OAI subscription tokens sold through resellers.) So I see most of the materialized risk here as still coming from Anthropic and OpenAI; safeguards aren't sufficient to stop a moderately dedicated actor.
The groups I know who are using open-weight models are legitimate offensive security companies. They won't break the rules to use subscription-based pricing, the open-weight models are more reliable in that they don't require specific jailbreaks nor hit classifiers, and the labs use massive partnerships or spend as a prereq for lowering classifiers/safeguards. I know of three legitimate groups running GLM 5.2 as their primary model.
That last part applies for Anthropic, too: I know of two instances where two different Anthropic GTM people used large comitted spend contracts as a prereq for lowering safeguards, and I directly witnessed one.
On the inside, I know the narrative and intent is genuinely about safety. But from the outside, Anthropic-the-system seems to be optimizing for revenue and control/power, isn't diffusing capabilities to defenders, and also doesn't have adequate safeguards to prevent misuse from dedicated bad actors.
As a result, I now lean towards a future where capable open models are freely available (at least for cyber, bio is harder); I don't trust Anthropic or other frontier labs to handle this sufficiently well without diffused capabilties given what I've seen so far.
