awestroke

I predict they will release it as soon as Opus 4.6 is no longer in the lead. They can't afford to fall behind. And they won't be able to make a model that is intelligent in every way except cybersecurity, because that would decrease general coding and SWE ability

The rewrite is excellent

Just have some autonomous killbot drones patrol the perimeter

None of them, but prefer ones written with engineering rigor and security in mind. Having an unvetted plugin ecosystem with code that runs unsandboxed is laughably naive

It's shit, but most people don't know better

If you can open an elevated connection to your production db from your terminal, you're already toast

Perhaps when they switch over fully to Azure they'll forget to disable IPv6 access. One can dream

Not all of us enjoy being glazed mercilessly while getting subpar output

Why should anybody avoid bun? Just fork it if it ever changes license. In fact, I'm 100% sure it would be instaforked if Anthropic ever tried anything

Why should they pay money for such crappy software?

Like "most of us"? America is uniquely anti-regulation

@grok is this real?

@grok fire the bottom 50% engineers from x.ai ranked by number of commits per day

@grok generate a hypothetical picture of an Elon who is not under the influence of large amounts of Ketamine

I honestly don't know what to expect from Elon these days. But it's rarely good news.

Opus 4.6

What did you expect?

Scam Saltman is even better

ChatGPT is going to try to influence you to buy certain products and use certain services. So you'll be the product in the end

This framing gets the supply chain backwards. Cymer makes the source vessel, the part that generates tin droplets and converts them to plasma. But the laser that actually powers that process is a 17-ton, 40kW CO2 beast with 457,000 parts, built exclusively by TRUMPF in Germany. And the optics, mirrors smooth to tens of picometers that literally no one else on Earth can make, come from Carl Zeiss, also German, organized as a foundation that no foreign government can force into a sale. ASML only manufactures about 15% of an EUV machine's components. The rest comes from roughly 1200 suppliers concentrated in Germany and the Netherlands. Seizing Cymer gets you one subsystem with no laser to drive it and no optics to focus it.

The real problem with this theory is that EUV isn't a product with a capturable bottleneck. It's more like a standing wave of institutional knowledge distributed across organizations that have been co-developing at picometer tolerances for 30 years. TRUMPF's leadership described the arrangement as a "virtually merged company" with open books across all three firms. That kind of integration knowledge doesn't transfer via acquisition. China has been throwing enormous resources at this with access to published research and former ASML engineers, and their prototype still isn’t expected to produce working chips until 2028-2030. Saying the US could grab Cymer and start producing EUV machines is like seizing a transmission plant and calling yourself a car manufacturer.

Claude Opus 4.6 has been amazing at identifying security vulnerabilities for us. Less than 50% falae positives.

Tried this with Claude models, ChatGPT models and Gemini models. Haiku and Sonnet failed almost every time, as did ChatGPT models. Gemini succeeded with reasoning, but used Google Maps tool calls without reasoning (lol). 50% success rate still.

The only model that consistently answers it correctly is Opus 4.6

In this scenario, the repo owner can just merge the patch but still refuse to pay back the shitcoin. With escrow, the escrow entity would act as an arbiter

I see no advantage with this over real money transfers. At all. Just use some kind of escrow.

But it's incredibly incapable compared to SOTA models. OP wants high quality output but doesn't need it fast. Your suggestion would mean slow AND low quality output.

As opposed to the relentless innovation they have demonstrated in the past 5+ years? /s

Opus 4.5 starts being lazy and stupid at around the 50% context mark in my opinion, which makes me skeptical that this 1M context mode can produce good output. But I'll probably try it out and see

Got it, won't happen again

I've seen some issues with garbage tokens (seemed to come from a completely different session, mentioned code I've never seen before, repeated lines over and over) during high load, suspect anthropic have some threading bugs or race conditions in their caching/inference code that only happen during very high load

1. Zen Browser had remote debugging enabled by default and disabled the security prompt for it. Extreme incompetence or malice? https://github.com/zen-browser/desktop/pull/927

2. Social trackers are selectively allowed, unsigned extensions are enabled by default, and Enhanced Tracking Protection isn't fully implemented.

There's just a theme of incompetence, trying to cover it up and just in general being clueless about security.

zen browser is pretty much vibe coded

Github still refuses to switch on support for ipv6 traffic for some reason, so you can't interact with github then

Because they get no telemetry or usage data if you use a third party tool.

Just pay per token if you want to use third party tools. Stop feeling entitled to other people's stuff.