pixelsort

8 ms·

31.
by pixelsort 2y ago
Working prompt injections for frontier models are devised by applying brilliant pattern constructions. If models ever become useful for writing them, that would represent a massive intelligence leap and a major concern. As things stand, wi
32.
by pixelsort 2y ago
What grift? I'm only reporting first-hand and second-hand anecdata -- some of which is observations from the "prompt whisperers" who follow in Pliny's circles. Chain of thought poses an existential risk to prompt injecti
33.
by pixelsort 2y ago
It isn't fundamental. As the models begin to leverage test time compute more effectively, prompt injection becomes more difficult. The models are becoming more sophisticated at detecting the patterns of gibberish intended to sow confus
34.
by pixelsort 2y ago
This reminds me of a similar idea I recently heard in podcast with Adam Brown. I'm unsure whether it is his original notion. The idea being, that if we can create AI that can derive special relativity (1905) from pre-Einstein books and
35.
by pixelsort 2y ago
They have been monitoring their GPT Store for emergent killer applications with a user base worth targeting. Zuckerberg's playbook. Nothing yet, because they've been too short-sighted to implement custom tokens and unbounded UIs.
36.
by pixelsort 2y ago
The hiring teams are employed. If they aren't in a position to fix the dynamics then nobody can. HR enjoys a vaulted position under which their suffering KPIs allow them to point their fingers at the market and shrug the blame from the
37.
by pixelsort 2y ago
Of course OpenAI can't stuff their UI with ads yet. In the middle of escalating anti-AI sentiment due to the rapid slopification of the infosphere that they intiated? With cultural resentment growing over the devaluation of visual evid
38.
by pixelsort 2y ago
Manager need to replace engineers with AI faster than engineers use AI to replace their managers. In the end, nobody wins except OpenAI.
39.
by pixelsort 2y ago
Thanks for digging that out. Yes, that makes sense to me as someone who made a fully local speech-2-speech prototype with Electron, including VAD and AEC. It was responsive but taxing. I had to use a mix of specialty models over onnx/w
40.
by pixelsort 2y ago
> You'll know AGI is here when the exercise of creating tasks that are easy for regular humans but hard for AI becomes simply impossible. No, we won't. All that will tell us is that the abilities of the humans who have attempte
41.
by pixelsort 2y ago
Any evidence potentially supporting that? To me it seems that every time a congress critter gets briefed behind closed doors they come out resolved to ban TikTok. Therefore, most likely, the Chinese probably have some insidious goal like pr
42.
by pixelsort 2y ago
Hmm, maybe send children with no primary attachment off to battle school?
43.
by pixelsort 2y ago
It works wonderfully until it fails miserably. At least, a couple years back that was the case. Someday, my local vision model will prefecth, take screenshots to check for my desired contrast, and iterate. Then the 3-5% of sites (perceived)
44.
by pixelsort 2y ago
We might have fewer future vigilante executive assassins if our government, and FTC in particular, would deem to enact consumer protections against things like AI-based medical claims recommendations that go beyond initial triage/routi
45.
by pixelsort 2y ago
Careful, we are not supposed to discuss how many human jobs will remain. The script clearly states the approved lines are "there will still be human jobs" and "focus on being adaptable". When it seems normal that Fortune
46.
by pixelsort 2y ago
All of the major AI providers are trying to pretend they care about your privacy by being weasles with their retention and anonymization terms. Partly why I'm building a zero-trust product that keeps all your AI artifacts encrypted at
47.
by pixelsort 2y ago
They have to, the major AI companies are ads companies. Their profits demand that we accept their attempts to normalize the Spyware that networked AI represents.
48.
by pixelsort 2y ago
For a detailed accounting of the challenges involved, I recommend this video: Why It Would Be Preferable To Colonize Titan Instead Of Mars https://youtu.be/_InuOf8u7e4?si=hRO1ZYCZtbQXUuK9 If you fully watch it, or already k
49.
by pixelsort 2y ago
As a counterpoint, she recently reviewed a paper in one of her recent videos and completely tore it to shreds as apparently the math was full of absolute nonsense. This was a fascinating watch, and not the kind of content that is easy to fi
50.
by pixelsort 2y ago
With respect to the American public, the majority can and do utilize nuanced thinking as a survival skill. The problem of modern American era, is not that our public is low in average intelligence. Rather, that on average, we have been mise
51.
by pixelsort 2y ago
Perception on the degree of severity seems to depend largely upon the size and structure of your social networks. In the post-GPT4 tech economy, your work friends are also the only worthwhile recruiters.
52.
by pixelsort 2y ago
The benefits would depends on how your code is structured, which dependencies you use, and overall complexity. Contexts provide and elegant developer experience, up until they don't. Have you ever implemented nested SplitView component
53.
by pixelsort 2y ago
There are 45 comments so far and 7 references to "redux". I'm just gonna rep for the Recoil clones here and mention Jotai, which likely already works with React 19 without any necessary changes. With enough front-end complexi
54.
by pixelsort 2y ago
It seems that anyone who has ever played games like Factorio or Satisfactory can readily extrapolate similar real-world conclusions. Physical inefficiencies are merely an interface issue that erodes over time with intelligent modularizatio
55.
by pixelsort 2y ago
Mach 11 hypersonic missiles? Is it even possible to intercept something going that fast without your interceptor tearing itself apart?
56.
by pixelsort 2y ago
Ha! Okay, I laughed at "computational phrenology".
57.
by pixelsort 2y ago
Could I get 1% as much for snugchat.com? Tempting, but I think I'd still rather build it out; especially now.
58.
by pixelsort 2y ago
Uh-oh. Academia is saying one of the quiet parts out loud.
59.
by pixelsort 2y ago
This is interesting; using NER for PII/secret filtering as bridge between secure systems and networked APIs. Over the last six months I built what seems to be the world's first desktop application providing a vault with encryption
60.
by pixelsort 2y ago
Most of it is mediocre creativity and a low willingness to adapt their prompting patterns for an optimal ratio of effort to output quality. Most people who don't understand yet expect LLMs to read their minds when they would better of

More ›