veunes

6 ms·

61.
by veunes 5mo ago
This is a good reminder that encrypted and end-to-end encrypted are very different promises
62.
by veunes 5mo ago
This is a lot less strange if you think of the robot as participating in ritual rather than "having faith"
63.
by veunes 5mo ago
45 is painfully young
64.
by veunes 5mo ago
The manipulation risk is real, but it usually comes from pretending there is a painless answer
65.
by veunes 5mo ago
The "old car" analogy seems right, with the extra complication that the car is supplying a non-trivial chunk of the country's electricity and replacing it is not quick
66.
by veunes 5mo ago
The interesting part will be whether Belgium can turn this into a coherent long-term plan
67.
by veunes 6mo ago
I bet if you could look at the hidden reasoning tokens at the exact moment the DB was dropped, there were zero thoughts about safety rules in there. The model simply hit an access error > searched for a token > found one > ran the
68.
by veunes 6mo ago
"backups in the same volume" aren't backups, they’re just snapshots in the same blast radius fwiw. If your DR plan hinges on a single physical volume ID, you have zero resilience This needs to be a lesson for everyone: real b
69.
by veunes 6mo ago
CORS protects your Facebook from your Gmail, but it won't protect your Gmail from the agent itself since it already has access to the DOM and JS context. If that agent gets hit with a prompt injection and decides to "Delete all ma
70.
by veunes 6mo ago
Sure, 26B models on beefy desktop silicon are finally nipping at the heels of commercial APIs, but this is a mobile thread. On a phone with 8GB of RAM and passive cooling, your tokens per second (t/s) are going to fall off a cliff afte
71.
by veunes 6mo ago
It’s likely a llama.cpp backend issue. On the Pixel, inference hits QNN or a well-optimized Vulkan path that distributes the SoC load properly. On the iPhone, everything is shoved through Metal, which maxes out the GPU immediately and cause
72.
by veunes 6mo ago
This article is all fluff because real benne marketing. If they mentioned that a 4B model on an iPhone 16 drains 15% of the battery for a single long prompt and triggers hard thermal throttling after 20 seconds, nobody would be clicking on
73.
by veunes 6mo ago
I noticed the inference is routed through the gpu rather than the Apple neural engine. Google’s engineers likely gave up on trying to compile custom attention kernels for Apple’s proprietary tensor blocks iirc. While Metal is predictable an
74.
by veunes 6mo ago
Yeah, a lot of it only becomes obvious in hindsight because each individual signal is easy to rationalize away
75.
by veunes 6mo ago
You're not just delivering expertise, you're stepping into a situation where incentives are already misaligned, expectations are fuzzy, and there's often a cashflow problem hiding somewhere
76.
by veunes 6mo ago
I think the hidden advantage here isn't even enforcement, it's filtering
77.
by veunes 6mo ago
This is all correct in principle, but in practice it's a lot messier
78.
by veunes 6mo ago
This reads less like a "got ripped off" story and more like a perfect storm of every classic consulting red flag showing up at once
79.
by veunes 6mo ago
Makes total sense. Consumer UX relies on pure determinism. When I click "Save", I know exactly what's going to happen. When I type a prompt into an "AI agent", I'm basically playing roulette every single time.
80.
by veunes 6mo ago
Because AI only drove down the cost of writing code, not the cost of finding Product-Market Fit. Sure, you can spin up another Notion or Jira clone over the weekend using Cursor or Claude Code now. But getting users to actually migrate thei
81.
by veunes 6mo ago
That’s exactly where we’re headed. Architecturally it makes zero sense to spin up an LLM in every app's userspace. Since we have dedicated NPUs and GPUs now, we need a unified system-level orchestrator to balance inference queues acros
82.
by veunes 6mo ago
It’s a neat idea, but giving a 2B model full JS execution privileges on a live page is a bit sketchy from a security standpoint. Plus, why tie inference to the browser lifecycle at all? If Chrome crashes or the tab gets discarded, your agen
83.
by veunes 6mo ago
Your whole premise is built on the OpenAI model, but that's not a moat, it's just a temporary API endpoint. The second a theoretical Llama 5 drops that's 10% cheaper and 5% smarter, every single startup in that "ecosyste
84.
by veunes 6mo ago
OpenAI overtaking Microsoft? Seriously? Microsoft has a massively diversified business spanning from gaming and cloud infra to B2B software that the entire world runs on. OpenAI has exactly one product (matrix weights), which is getting hea
85.
by veunes 6mo ago
Demand is stagnating only applies to the B2C segment, where people are already bored of generating poems and funny pictures. In B2B, the demand hasn't even started yet because corporations are still terrified of shoving their NDA data
86.
by veunes 6mo ago
Bingo. Even if some magic drops tomorrow that compresses the KV cache down to literally zero bits, that saved VRAM will instantly get swallowed up by bumping the batch size or pushing the context window to 10 million tokens. There is no suc
87.
by veunes 7mo ago
You can ask someone to take off glasses or power down a phone, but you can't really "check" an implant in the same way
88.
by veunes 7mo ago
A fixed security camera is visible, scoped to a specific area, and (in theory) governed by policy, retention limits, audits, etc. Smart glasses are much harder to notice and can record anywhere
89.
by veunes 7mo ago
Yet I get why courts are nervous about anything with a hidden camera/mic
90.
by veunes 7mo ago
Yeah, "we promise not to use it" is about the weakest possible control in a situation like that

More ›