HNHacker News
TopNewBestAskShowJobs

make3

1,914 karma · joined April 15, 2013

FFEEDD ff6600 a77858
submissionscomments
make3··on The text in Claude Code’s “Extended Thinking” output
this prevents you from using any commercial model then, because commercial models need to hide thoughts to prevent distillation
make3··on Windows 11 New Media Player Uses 3.5x More RAM, Charges for Popular Video Codecs
OP was clearly a joke about Copilot making bad code
make3··on Claude Fable 5
very repetitive, all sorts of weird style decisions & reimplementions for no reason
make3··on Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?
Claude Opus 4.0. ok but no one cares about that old version man
make3··on AI Engineer Claims to Have Cracked Linear A
I'd like to push back
make3··on GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2
Transformers are not Markovian, their whole point is arguably to be the reverse of Markovian, to efficiently make it so the new tokens are a function of all previous tokens
make3··on John Jumper to join Anthropic
Shazeer is mostly an IC, weirdly enough (source: I worked with him).

Demis is the CEO of DeepMind, it's completely different.

Jumper.. the AlphaFold team left & made Isomorphic. I was always surprised that Jumper hadn't gone with them.

make3··on John Jumper to join Anthropic
I just think that people are running to pre IPO Anthropic to make bank.

It also feels like Anthropic is the new Google though. They actually try to not be evil, and are actually at the frontier of new tech.

make3··on John Jumper to join Anthropic
you can't know if an improvement is just due to compute, this was a lesson of the early scaling, the whole emergence phenomena, new capabilities emerge even only from more compute & data
make3··on John Jumper to join Anthropic
in theory the IPO pricing is already set
make3··on Has AI already killed self-help nonfiction books?
This felt like a vacuous post by a LinkedIn influencer.
make3··on Improvement in advanced Alzheimer’s disease following high-dose psilocybin
Pharmaceuticals are not a question of taste like alcohol and food, they're strictly a matter of purity, stability and toxicity, which is clearly better supported by being produced by an industrialized country with sometimes hundreds of career chemical engineers running the show in a billion-dollar company with processes etc., & who have liability for lawsuits worth hundreds of millions of dollars if they put anything remotely close to the shit recreational drugs are cut with, like horse tranqualizer and deworming agents and whatnot.

The best elites can dream of is pharmaceutical levels, but even the most pure illegal stuff comes from a production line that's isolated from the rule of law and from transparency and liability and also likely from actual professionals, so it's likely still more toxic than it needs to.

The real "advantage" of cocaine is mostly that the molecule has an inherently stronger euphoric effect, I would assume (I've never taken it), also making it a worse therapeutic if you want to actually do a better job vs just feel "high".

make3··on Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?
There is no Claude 4 Opus model... It's a series of model, of which the strongest is Opus 4.8, and Qwen 3.6 35B-A3b gets 51.5% on Swe-bench pro to Opus 4.8's 69.2%
make3··on Improvement in advanced Alzheimer’s disease following high-dose psilocybin
why take cocaine when you can get your doctor to prescribe Adderall to you
make3··on Improvement in advanced Alzheimer’s disease following high-dose psilocybin
hallucinogens where taken by hippies, it was probably a way to put a few extra commies and lefties in prison
make3··on CRISPR tech selectively shreds cancer cells, including "undruggable" cancers
it's revolutionary in genetic research, allows us to observe cells with specific gene knockouts, over activations, etc. it's just very hard to deliver to a whole system
make3··on AI agent bankrupted their operator while trying to scan DN42
maybe de-rlhf unleashed agents
make3··on Chrome is looking to permanently drop MV2 extension
Will ublock work with this?
make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
you don't get the model when you buy the data center, & no amount of running smaller models on a tiny 200k$ "cluster" (that's like one 4 gpus node, not even 8) will get you remotely close to Fable 5 level performance
make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
the feedback is for something mindless though, "we don't care about societal harms". I wonder the overlap between these commenters and tech maga people, eg crypto bros & Elon stans.
make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
You're being too literal, they're saying people are not thinking with a philosophically interested mind, which is blatantly the case here, their point stands.
make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
"I will push back and say" it's only paternalism if it's about helping the user's not harm themselves.

This is about societal impacts, not wanting their models to be used by some people against other people, as a weapon.

make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
this reasoning is inverted lol they would get a lot more information by letting you use it. so much weird drama around reasonable guardrails for an experimental model
make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
Why is this surprising or a problem?! It's a model demo, & their reasoning is reasonable and fair. Why all this drama.
make3··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
what does this mean
make3··on Claude Fable 5
4.6 had functional but very poor quality code
make3··on Anti-social: It's fads, not friends, which now dominate social media feeds
parasocial media
make3··on I design with Claude more than Figma now
I think it's normal for non-technical people to have no idea of how things work and of what to expect. It's our responsibility to communicate. And, within reason, theirs to listen.
make3··on I design with Claude more than Figma now
they don't know that it'll break, that's the problem of course
make3··on Anthropic, please ship an official Claude Desktop for Linux
I also care about my desktop supporting consumer electronics without having to waste an afternoon debugging Bluetooth drivers every two weeks
← PreviousPage 5 of 34Next →