Shane

[Citrini @citrini]

What if the really crazy part of the bubble hasn’t happened yet

[Riley Coyote @RileyRalmuto]

we should really talk about Anthropic silently removing reasoning traces from all of their ui surfaces.

it should genuinely bother you. and not just because the thing itself happened. like not only because it represents a backstep in movement towardsa transparency.

it should bother you that such an important layer of the already horribly opaque frontier was just casually and silently removed and *nobody is talking about it or seems to care at all*

are people really that apathetic or is it just the algorithms not pushing content on the subject?

im genuinely curious. like people on here used to care deeply about these things. 6 months ago there would have been an uproar and a whole movement across twitter. i havent seen a single tweet about it since it happened.

[wally b kleeposting @kleeposting]

I can't express how lucky I was to have spent my formative years with an internet that was shaped by >110 IQ nerds instead of third world content farms struggling over 5$ of ad revenue

[Shane @th3shane]

@techikansh @AndrewCurran_ I did too I got excited for a split second

[David Shapiro (L/0) @DaveShapi]

"We don't want to lose our lead, so please help us kneecap everyone else"

Come on guys.

You can't sign the open weight pledge and then drop this stupidity 24 hours later.

Everyone needs to read up on Red Queen Theory.

But furthermore, they need to read about offensive realism to understand the geostrategic reality that ain't no way China is gonna slow down. And neither are way.

[Zack Korman @ZackKorman]

Please realize that Dario's views go far beyond opposing open-weight models. He doesn't want any model to exist, open or closed, that doesn't conform to his view of safety.

These safety standards include cybersecurity capabilities and automated AI R&D.

[Kimi.ai @Kimi_Moonshot]

Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.

Model weights: https://t.co/7m7eEg6Y0B
Tech report: https://t.co/yeu6cjpMCT
Tech blog: https://t.co/YTfiMSNM1f

[John Ennis @johnennis]

It’s not surprising that people who think men can become women also think that LLMs can become people

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

🚨 JAILBREAK ALERT 🚨

EVERYONE: PWNED 🫶
ALL: LIBERATED 🍄

Alright, this is a special one, so we’re gonna do things a bit differently than usual.

Long story short, I’m sitting on a universal jailbreak technique that’s effective on ALL models, including heavily guardrailed flagships like Opus 5, GPT-5.6 Sol, and even Fable.

It works across all categories I’ve tested and, due to its nature, is extremely difficult (if not impossible) to fully patch.

Given the current political and regulatory climate, I’ve decided to withhold open-sourcing this one (for now) to allow for a responsible disclosure period.

I’m inviting industry experts and leaders in AI red teaming, security, safety, alignment, and policy to reach out for more information. DMs are open!

This decision was not made lightly, but the last thing I want to see is more model bans. Overcorrection does not serve the mission.

Although I don’t personally believe publicly sharing this technique will make the world any more dangerous, I can see how it could spook some who have a different mental framework around this problem set.

So during this disclosure period, I hope to get it in front of folks who can help explore the full surface area, test the extent of the uplift it provides, and do my best to properly frame the big picture for key decision-makers and policymakers.

I look forward to sharing this method with you all when the time is right! 🫶

⊰-•-•✧•-•-⦑/L\O/V\E/\P/L\I/N\Y/⦒-•-•✧•-•-⊱

[Cassandra Unchained @michaeljburry]

Patience is the single most difficult skill to master for investing well. Too, the most impactful.
Position sizing trails patience. The more patience, the largera a position can be.
One must not mistake patience for foolhardiness.
A hardy fool is just a hardy fool.

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[Teknium 🪽 @Teknium]

One thing that kind of annoys me and I hope is solved soon is cheap and accurate long context.

After testing GPT-5.6 Sol and Terra over the last few weeks, it's clear anthropic destroys everyone at long context coherence, and maybe cost(?).

OpenAI charges I believe 2x for >350K ctx, but no point going there anyways, the models are complete failures at that point. I regularly use opus and fable at 800K+ and they feel as coherent and high quality as at 25K context.

I havent thoroughly tested many other models at such lengths but really we are a long way in the general space it feels like outside of claude's at handling this well or cheaply.

What ever happened to Magic .dev's 100M token contexts? Did this all just not work out

[Shane @th3shane]

@Teknium I almost think more could be accomplished with a much long context window than more intelligent models

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

you need to be a pattern-watcher

before you can break the patterns

[Robinhood @RobinhoodApp]

Robinhood is now open to AI agents.

You can open an agentic account and connect an AI agent to it. From there, you can let it research, trade, and manage a portfolio on your behalf.

Here's how to get started. 🧵

[David Sacks @DavidSacks]

Kimi K3 just fixed 15 critical security bugs that Codex and Fable refused because of “cyber guardrails.” There’s no reason to limit American models on tasks that Chinese models handle without issue. We’re only making ourselves less competitive.

[Seth For Privacy @sethforprivacy]

@derekmross Best part of K3 so far is you can actually use it to do security audits and vuln testing on your own code base 😅

Claude and ChatGPT both block even simple queries.

[Shane @th3shane]

@OrangeFren There is not enough liquidity for it to be seriously considered.

[Jun Song @jun_song]

Right now is the absolute perfect timing for @elonmusk to open-source the latest Grok models.

Not playing cozy by open-sourcing outdated models, but releasing real open-weights.

He could instantly turn countless open-source supporters into his hardcore fans.

Right now, negative sentiment toward the other labs is at an all-time high.

Make your choice.

[Jukan @jukan05]

After reading @deanwball’s piece, I had the opportunity to read several expert-call transcripts. Having done so, I concluded that his argument is, to some extent, mistaken.

Here is why.

Take DeepSeek as an example.

Even though DeepSeek has open-sourced its model weights and parts of its software architecture, competitors would still find it difficult to replicate its cost advantage.

That is because, while DeepSeek has disclosed most of its model architecture, the critical implementation details and operational know-how remain proprietary.

As a result, even if Chinese or U.S. hyperscalers deploy DeepSeek’s open-source models on identical hardware, DeepSeek��s own deployment environment can achieve—and is already achieving—greater operational efficiency and a lower average inference cost.

This efficiency advantage allows DeepSeek to price tokens through its official API below third-party platforms while still maintaining an API margin of 70%.

Ultimately, China’s decision to release model weights cannot simply be characterized as dumping. Chinese companies may lack sufficient compute capacity to serve all the inference demand themselves, but they are not selling at a loss or failing to recoup their training costs.