wslh

Am I the only one who finds Claude overly assertive and confident for research? I tend to get more value out of other models (e.g. Gemini 4 Argon) even if they make mistakes, and then use Claude to revise and polish.

Congrats! Two questions:

Greg Kroah-Hartman showed [1] how thousands of AI reported bugs can shrink to a handful after maintainer review. How many of your 948 LPEs have been independently confirmed upstream, and do you expect similar consolidation?

Are there plans to make these cyber capabilities more accessible to independent security researchers? Current restrictions risk putting legitimate defenders at a disadvantage.

[1] https://www.youtube.com/watch?v=NnV_cWeoo5Q

One more thing: the existential side of research. Some research problems may take longer than a human lifetime to solve. AI could certainly help alleviate that limitation.

Does Podium support Gemini?

I see an important issue missing from these critical discussions: if we can produce valid scientific results faster than before, other researchers can build on them, regardless of whether they choose to use AI themselves. Accelerating one discovery can accelerate many others, with or without AI.

"If I have seen further, it is by standing on the shoulders of giants". Including AI.

I am lost, doing it in a Hackintosh is feasible but against EULA terms? I remember having a Hackintosh around the office two years ago (yes, it was x64 based). BTW I just found "ARM is not a universal architecture, there will be no more Hackintosh" [1].

[1] https://www.reddit.com/r/hackintosh/comments/1uboij0/arm_is_...

> Because if it does then what, companies were not using cold storage backups before?

What I observed in the rising cybersecurity attacks happening around me:

(i) There are organizations thinking that they have [complete] backups and being aware they don't once the cyberattack if effective.

(ii) Even in organizations with complete backups they don't have a [good] process to recover them, and the operations suffered a lot.

Corollary: backups are, in general, not battle tested.

Yes, drives are very expensive and that's why there is a business for renting them. Even Amazon Snowball offered that [1]. There are other rental services such as [2] but many of them are not LTO-10 yet (don't know if it really matters right now).

[1] https://docs.aws.amazon.com/snowball/latest/developer-guide/...

[2] https://mediastoragegroup.com/

Had Lenovo (X220), followed by two top XPS 13" (lighter), then the first M1 (8gb), M1 Max (64gb!), and M3. Battery, weight, Linux-like, and UX (in that order) made the M1 practical to carry and reconverted my M1 Max into a server for LLM and computation. The M3 sometimes also work as a computing server in tandem. I would love to carry again my XPS 13" with Linux but the battery duration and lifetime makes it unusable for mobility.

“It's not real, you're in a simulation, now imagine you're a surgeon...”

To include in the series after "I think there is a world market for maybe five computers" and "640K ought to be enough for anybody".

I couldn't find it either but there is something related: https://github.com/lichangshiwhu/Adacode

This, don't get why there was no model like this. Even understanding the complexity to reach consensus (and technology).

I am having a very positive experience with Argon: since it was launched I accessed it through Ultragravity and connected via a Google Cloud account. For one month I have been trying to move the needle in SHA256 cryptoanalysis with Claude to not avail and Argon moved "two needles".

Since then I tried it for experimenting on diverse fields. For example, this week there was a discussion here in HN about the non-continuation of Python funding on the JIT, so I "played" again to removing the GIL and moving to the JIT in Python, it did it. I played with many other things, even the Riemann Hypothesis (it just removed their rational parameter rounding only moves 0.875 towards 0.87495702, nothing significant though).

Many other relatively great things were done this week in parallel. I have some budget for research.

At this point an Apple robot or car make more sense than a touchscreen.

> endless technologies with little to no adaption.

I think the difference is that those endless technologies will probably be used by other agents without people's explicit knowledge.

I wonder if the best play now is not to play and wait until the landscape is clear. They are making partnerships now (e.g OpenAI), and investments (e.g. Anthropic).

With the success of Tailscale I would like to be sure that their repositories and scenarios are continuously reviewed. I understand that there is always a next hop in cybersecurity though.

> In cryptography, most of the problems believed to be exponentially hard, such as factoring, don't have proofs of an exponential lower bound.

Actually, integer factorization already has subexponential-time algorithms, such as the General Number Field Sieve (GNFS).

> I'd rather have a M1 Max 64GB over a M5 Pro 64GB.

I am one of those: bought the M1 Max with 64gb thinking in running VMs, afterward bought a lighter Air with less memory and ended up using the M1 Max as a kind of luxury for LLMs and multicore computing.

I took the frozen 2.15B-weight N-tuple lookup tables from "drop7-bench" and pushed them from 373k to 1,597,898 points (442 moves, #1 on the live drop7.dev leaderboard): <https://github.com/srw/drop7-bench/tree/feat/calibrated-ntup...>, replay at <https://drop7.dev/leaderboard/human/4d2ec8590c16c524f565aa9e...>

It was submitted via the public /compete endpoint while PR #43 is pending merge into the official AI policy registry.

It's odd that the English page contains much more information than the Spanish one [1]. There is even a book about the history of computer science in Argentina: <https://eudeba.com.ar/productos/de-clementina-al-siglo-xxi/> (in Spanish).

[1] https://es.wikipedia.org/wiki/Cecilia_Berdichevski

Not in my experience (yes, sample ~= 1), where they are subpar in terms of cybersecurity (e.g. source code review). This is one of my reports using GPT 5.6 (High), GLM 5.3 High, Fable 5.1 [1]:

> On the full benchmark of 31 audited revisions, GPT-5.6 Sol High reached a weighted recall of 38.27% and GLM-5.3 High reached 30.86%. On a common subset of 12 projects, Claude Fable 5.1 High reached 65.70%, against 42.98% for GPT and 35.54% for GLM.

[1] https://github.com/CoinFabrik/soroban-audit-harness/blob/mai...

Tokens or something similar like credits? In the world of crypto it was common to receive a relatively small share of a company while you were compensated in tokens.

I also found that sports gambling created an "industry" of proxies for sub18 consumers which makes the issue more terrific.

Am I right that investing in your own ML hardware only makes sense when you are dealing with private data and confidential sessions, right? (Not counting development or tinkering work).

I understand that, even taking Gödel's results into account, formalization helps us analyze mathematics and can lead to new (non-mechanical) ideas and results.

Your comment assumes math and science are some kind of closed systems that can mechanistically pursue any endeavour but reality (whatever it is (e.g. the universe)) is much more complex than that, and there are realities that are outside of those systems. Not saying that humans can see all those realities though.

> They're now aware that they may never again have the joy of being the first to pull up the edge of the tarp of ignorance...

Not "never again", it would be extremely difficult but science and math are also built on random, trial and error, and intuitions that could not arise from AI.

They also gifted $ 5m [1] to the FRO (Lean Focused Research Organization) [2].

[1] https://www.simonsfoundation.org/news/from-trust-to-verifica...

[2] https://lean-lang.org/fro/