Hacker News: dogacel

Framing an LLM as a safety researcher changes its language, not its judgement

dogacel — Sun, 08 Feb 2026 05:29:31 +0000

Article URL: https://lab.fukami.eu/LLMAAJ

Comments URL: https://news.ycombinator.com/item?id=46931581

Points: 1

# Comments: 0

Training language models on TPUs shouldn't be scary

dogacel — Thu, 05 Feb 2026 16:20:16 +0000

Article URL: https://dogac.dev/blog/2026/migrating-to-tpu/

Comments URL: https://news.ycombinator.com/item?id=46901302

Points: 3

# Comments: 0

Show HN: Universal DeepSeek OCR 2 – CPU, MPS, CUDA Support

dogacel — Wed, 28 Jan 2026 15:31:09 +0000

Article URL: https://github.com/Dogacel/Universal-DeepSeek-OCR-2

Comments URL: https://news.ycombinator.com/item?id=46796616

Points: 2

# Comments: 0

Show HN: DeepSeek-OCR with MPS and CPU Support

dogacel — Fri, 21 Nov 2025 16:49:04 +0000

Article URL: https://twitter.com/dogacel0/status/1991629702575079783

Comments URL: https://news.ycombinator.com/item?id=46006139

Points: 3

# Comments: 0

AI can do a better job of persuading people than we do

dogacel — Fri, 23 May 2025 04:25:45 +0000

Article URL: https://www.technologyreview.com/2025/05/19/1116779/ai-can-do-a-better-job-of-persuading-people-than-we-do/

Comments URL: https://news.ycombinator.com/item?id=44069844

Points: 3

# Comments: 0

New comment by dogacel in "Anthropic CEO claims AI models hallucinate less than humans"

dogacel — Fri, 23 May 2025 04:19:48 +0000

I think the title should change as

"AI models hallucinate less than AI company executives"

New comment by dogacel in "Temporal: Open-source durable execution system"

dogacel — Wed, 14 May 2025 14:23:51 +0000

I recently discovered them, it is pretty cool. I really wish we could use it for our business workflows.

Also check https://conductor-oss.org/

Rethinking Modern Asynchronous Paradigms

dogacel — Wed, 14 May 2025 14:19:35 +0000

Article URL: https://blog.dogac.dev/modern-asynchronous-paradigms/

Comments URL: https://news.ycombinator.com/item?id=43984881

Points: 1

# Comments: 0

Writing at the Speed of Thought [video]

dogacel — Mon, 12 May 2025 06:16:36 +0000

Article URL: https://www.youtube.com/watch?v=sqm4-B07LsE

Comments URL: https://news.ycombinator.com/item?id=43960137

Points: 1

# Comments: 0

New comment by dogacel in "Semantic Unit Testing"

dogacel — Sun, 20 Apr 2025 23:56:46 +0000

What is the expected cost of running those tests? I think it would be a good argument to compare this to running cost of unit / integration tests for a function (based on action runners).

If it is so expensive, it can be distributed as a simple GitHub integration / action that runs perodically, what do you think?

Also see, https://dl.acm.org/doi/abs/10.1145/3663529.3664458

Adopting SLI/SLO for improving reliability

dogacel — Sun, 20 Apr 2025 23:46:55 +0000

Article URL: https://techblog.lycorp.co.jp/en/sli-and-slo-for-improving-reliability-2

Comments URL: https://news.ycombinator.com/item?id=43747398

Points: 2

# Comments: 0

What to Write (2024)

dogacel — Sun, 20 Apr 2025 21:32:29 +0000

Article URL: https://endler.dev/2024/what-to-write/

Comments URL: https://news.ycombinator.com/item?id=43746673

Points: 4

# Comments: 0

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Fri, 18 Apr 2025 20:21:03 +0000

> A counter that can be synchronized on demand is kinda superfluous --- not really secret and not terribly relevant either. All else being equal, an attacker can sync up just as easily as a legitimate client so why bother with the counter?

A unique counter for each authorization attempt ensures the resulting key is different for each attempt, which makes replay attacks not possible. I agree if you sync the counter two ways, it is better to use a "nonce", a totally random secret each time.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Fri, 18 Apr 2025 01:41:14 +0000

Why counter is the shared secret? In TOTP time is the counter and it is obviously not secret, so there is no reason to think the counter would be secret as well. Clients can sync their counter to match the server.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Tue, 15 Apr 2025 22:47:03 +0000

Some banks in Switzerland give customers a device that generates TOTP codes.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Tue, 15 Apr 2025 18:00:43 +0000

No, RSA is asymetric, where it has a public/private key pair.

HMAC is symetric, it only has a secret and it can be used to hash values one-way.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Tue, 15 Apr 2025 17:58:46 +0000

Agree and disagree,

Deciding on how to store the credentials is still a hard task. Even storing the secret. Ideally it shouldn't stay as a plain text in your database. If you use cloud, something like KMS can be used for additional security. Also you should still consider replay attacks, rate limits etc.

I agree in the sense that TOTP is hard to implement, no it is not. I hope this article helped people understand how TOTP works.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Tue, 15 Apr 2025 17:55:17 +0000

Hashing is done before storing the secret on the server side. Therefore they still need to communicate regarding the intial secret.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Tue, 15 Apr 2025 14:58:53 +0000

Even though QR codes are standardized, the original RFCs do not use QR codes. That's what I tried to mean, you can't find apps that use plain-text secrets.

New comment by dogacel in "Behind the 6-digit code: Building HOTP and TOTP from scratch"

dogacel — Tue, 15 Apr 2025 14:13:27 +0000

I also don't use email subscriptions, unless it is my favorite person. But I don't know how many of them are subscribed via RSS.