toraway

15 ms·

181.
by toraway 8mo ago
You can make the argument that the boat variant is ambiguous (but a stretch), it's really not relevant since the point was revealing the underlying failure mode is unchanged, just concealed now. The original car question is not ambiguo
182.
by toraway 8mo ago
That problem went viral weeks ago so is no longer a valid test. At the time it was consistently tripping up all the SOTA models at least 50% of the time (you also have to use a sample > 1 given huge variation from even the exact same wor
183.
by toraway 8mo ago
In reality: millions of boomers are scrolling FB this very minute reacting to the most obviously fake rage/surprise/love bait AI slop you've ever seen.
184.
by toraway 8mo ago
Hah, it's somewhat ironic how this is almost the exact opposite of the prevailing folk wisdom I've read for the last 1-2 years: that you should never use negative instructions with specific details because it overweights the exact
185.
by toraway 8mo ago
Latin/Greek were considered part of the core curriculum for a well-rounded classical education in the upper-class for hundreds of years (some degree of retained proficiency wasn't unusual in graduates of the elite schools in Brita
186.
by toraway 8mo ago
Yeah, it seems "sandboxing" is the current catch-all buzzword in AI products to hand-wave away any security concerns. Which often raises more questions than it answers for something like a generalist dev agent that has access to a
187.
by toraway 8mo ago
Yep, I see both Codex and Opus routinely circumvent security restrictions without skipping a beat (or bothering to ask for permission/clarification). Usually after a brief, extremely half-hearted ethical self-debate that ends with &quo
188.
by toraway 8mo ago
The parent comment specified a company that has never turned a profit cornering the entire market.
189.
by toraway 8mo ago
It's a neat idea but it's not exactly plausible real world conditions to have an agent that pretty much exclusively spends its time wading through an email inbox that's 99% repeated prompt injection attempts. As the creator a
190.
by toraway 8mo ago
It's classic hype/FOMO posturing.
191.
by toraway 8mo ago
The more straightforward explanation for the original OP's question is that they realized what they were doing was reckless and given enough time was likely to blow up in their face. They didn't hide because of a vague fear of bei
192.
by toraway 8mo ago
Jeff Dean literally featured it in a tweet announcing the model. Personally it feels absurd to believe they've put absolutely no thought into optimizing this type of SVG output given the disproportionate amount of attention devoted to
193.
by toraway 8mo ago
For a personal plan to use premium Gemini AI features or for agentic development with Gemini CLI/Antigravity the billing is no more or less complicated then Claude Code or Codex CLI. You pay for the $20/mo Google AI Pro plan with
194.
by toraway 8mo ago
But that idealized example could also be offset by another employee who doubles their own output by churning out lower-quality unreviewed workslop all day without checking anything, while wasting other people's time.
195.
by toraway 8mo ago
The usage limits for Codex CLI vs Claude Code aren't even in the same universe. Maybe it's not a problem on the web, but I never use the actual chatbots so I have no idea tbh. You get vastly more usage at highest reasoning level
196.
by toraway 8mo ago
Hah, your experience is a great example of the futility of recommendations to add instructions to "solve" issues like sycophancy, just trading one form of insufferable chatbot for something even more insufferable. Different stroke
197.
by toraway 8mo ago
Per another of OP's comments in this thread, they seem to suggest that CBS (of all organizations) is actually trying to help a Democrat with better chances in the general to defeat a Republican. So, therefore, they artificially manufac
198.
by toraway 8mo ago
Uh, why exactly are we inventing a completely speculative alternate possibility when this is perfectly in line with Brendan Carr's multiple public statements and recent examples wielding FCC regulatory power to strong arm media organi
199.
by toraway 8mo ago
"You guys are reacting like engineers" is a very wave-y dismissal of the many practical questions raised about why exactly OpenAI should expand into a product that's tangentially related (at best) to their core competency o
200.
by toraway 8mo ago
The meme about AMA artificially limiting the supply of new residency slots turned out to be much more complex the more I read about it (and IMO mostly incorrect, at least as described by the just-so stories on the internet where I first lea
201.
by toraway 8mo ago
But Anthropic very much is subsidized by VC money, just like OpenAI. They just raised another $30 billion this week . I'm not sure how Anthropic managed to position themselves as the "good guy" in many people's minds,
202.
by toraway 8mo ago
That is correct. I "came of age" in 2010-11 during the Web 2.0 era of web apps beginning to eat the world. Ruby was just starting to come down from its peak as the new hotness, and Python thanks to Django + Google's support&#
203.
by toraway 8mo ago
Yes, I'm entitled because I didn't stick around paying for a subpar plan compared to their direct competitor OpenAI who supports my use case at the same price point. Any reasonable person should be thanking Dario for lock-in tha
204.
by toraway 8mo ago
Okay, but why on earth should I as an OpenCode user accept that limitation when OpenAI explicitly supports 3rd party clients? That's how competition works in a healthy market. I certainly haven't built up enough brand loyalty to
205.
by toraway 8mo ago
I generally lean towards skeptical/cynical when it comes to AI hype especially whenever "emergence" or similar claims are made credulously without due appreciation towards the prompting that led to an outcome. But based on my
206.
by toraway 8mo ago
Huh? AI labs are routinely spending millions to billions to various 3rd party contractors specializing in creating/labeling/verifying specialized content for pre/post-training. This would just be one more checkbox buried in h
207.
by toraway 8mo ago
So, you've said multiple times in the past that you're not concerned about AI labs training for this specific test because if they did, it would be so obviously incongruous that you'd easily spot the manipulation and call the
208.
by toraway 8mo ago
I use agentic tools daily and SOTA models have certainly improved a lot in the last year. But still in a linear, "they don't light my repo on fire as often when they get a confusing compiler error" kind of way, not a "I
209.
by toraway 8mo ago
IMO, an extreme outlier in a system that was still fundamentally dependent on learning to develop until suffering from a defect (via deterioration, not flipping a switch turning off every neuron's memory/learning capability or som
210.
by toraway 8mo ago
Your suggestion would have been well taken and helped contribute to an elevated level of discussion characteristic of HN if you had instead simply pointed out the better source.

More ›