HNHacker News
TopNewBestAskShowJobs

pr337h4m

4,486 karma · joined August 20, 2022

preetham<zero><four><zero><eight> @ the largest mail provider
submissionscomments
pr337h4m··on OpenAI Discusses Giving Altman 7% Stake in For-Profit Shift
Microsoft is extremely likely to get rugged:

“Fifth, the board determines when we've attained AGI. Again, by AGI we mean a highly autonomous system that outperforms humans at most economically valuable work. Such a system is excluded from IP licenses and other commercial terms with Microsoft, which only apply to pre-AGI technology.”

https://openai.com/our-structure/

pr337h4m··on Did ChatGPT just message me first?
This is real, link to the actual chat: https://chatgpt.com/share/66e718e5-c934-8001-8ffa-ef7ca1f165...
pr337h4m··on SpaceX update regarding Starship FAA flight approval
What, if anything, prevents SpaceX from using offshore launch pads? What about launches from international waters?
pr337h4m··on Brazilian court orders suspension of X
Nah, the Brazilian government will lose this round. X is not going to cave, and the government doesn’t have the technological capacity to quickly spin up their own Great Firewall.
pr337h4m··on Starlink will now be free in Brazil since remote hospitals, schools use it: Musk
some context: https://www.americasquarterly.org/article/brazils-most-power...
pr337h4m··on SearchGPT Prototype
https://github.com/ItzCrazyKns/Perplexica https://github.com/miurla/morphic

LLM frontends for search engines have been commoditized for a while now, this doesn't seem to add anything new (at least technologically)

pr337h4m··on OpenAI Acquires Rockset
Looks like we'll finally be able to search our past ChatGPT convos without having to Ctrl-F the data export
pr337h4m··on Anonymous Source Shared Leaked Google Search API Documents
https://github.com/googleapis/elixir-google-api/commit/078b4...

https://hexdocs.pm/google_api_content_warehouse/0.4.0/api-re...

pr337h4m··on LLMs are not suitable for brainstorming
A simple "Be brutally honest" at the end works surprisingly well.
pr337h4m··on Misconceptions about SB 1047
Asterisk is tied to the effective altruism movement, that has been spending a lot of money lobbying for AI regulation: https://forum.effectivealtruism.org/posts/Mts84Mv5cFHRYBBA8/...
pr337h4m··on Welcome to Scam World
This is actually pretty easy to do with local LLMs: https://justine.lol/matmul/#:~:text=For%20example%2C%20let%2....
pr337h4m··on TikTok CEO expects to defeat US ban: 'We aren't going anywhere'
The text of the bill literally mentioned "ByteDance Inc" and "TikTok" by name.

(which makes it a bill of attainder, and thus unconstitutional)

pr337h4m··on Apple's risky bet on CarPlay
Why not use plus codes? They're well-integrated into Google Maps: https://maps.google.com/pluscodes/
pr337h4m··on Grok-1.5 Vision Preview
Looks like Optimus is an intended use case.
pr337h4m··on GPT-4 Turbo with Vision is a step backwards for coding
How hard could it be to let ChatGPT Plus users choose model versions? (especially when older versions are accessible through the API)
pr337h4m··on GPT-4's Hidden Cost: Is Your Language Pricing You Out of AI Innovation? (2023)
This is quite significant if you're trying build LLM-powered apps that are meant to be used in a multilingual context.

However, this is not really OpenAI's fault: the English writing system is much more simple compared to those of Chinese or Hindi, and even languages like French, Spanish and German, that are written in the same Latin script as English, make much more use of diacritics compared to English. So it is natural for the tokenizer to be much more efficient for English compared to languages with more complex writing systems.

pr337h4m··on Canadian 'online harms' bill to make online hate punishable up to life in prison
Text of the bill: https://www.parl.ca/DocumentViewer/en/44-1/bill/C-63/first-r...
pr337h4m··on Netlify just sent me a $104k bill for a simple static site
Any cloud service that offers prepaid billing on a credits model would get a lot of customers.

(Prepaid is also superior from a cash flow perspective too.)

pr337h4m··on Argentina's Economy Is Screaming at Top of Its Lungs
I wouldn't consider myself to be knowledgeable about Argentina, but this is a blatantly misleading article.

All the conclusions here are based on the government's official USD/peso exchange rate, which until a recent devaluation, had been completely out-of-sync with the black market exchange rate.

From this AP article published in December: https://apnews.com/article/argentina-inflation-milei-currenc...

"The previous government sought to deny reality by strictly limiting Argentinians’ ability to exchange pesos into U.S. dollars or other foreign currencies. As a result, the official exchange rate made the peso look stronger than it actually was — around 400 pesos for every U.S. $1 before the devaluation that Milei’s government announced Tuesday. But no one was fooled. The black market has lately pegged the peso at around 1,000 per $1."

"At the heart of Milei’s audacious economic agenda is his plan to devalue the peso from 400 to 800 per $1 U.S. dollar and then by an additional 2% each month. Part of the goal is to make Argentina’s exports less expensive — and thus more competitive — overseas and reduce the country’s gaping trade deficit."

"Some economists worry that Milei’s devaluation doesn’t actually go far enough. His plan would narrow — but not close – the gap between the official exchange rate and the 1,000-peso-to-$1 rate in the black market."

pr337h4m··on Show HN: Supermaven, the first code completion tool with 300k token context
Off topic, but do you have any plans for same.energy? No image search tool has surpassed it even after two years.
pr337h4m··on Bluesky announces data federation for self hosters
The whole point of end-to-end encryption is that you don't have to trust the people running the service you use.

If Signal releases a malicious update (and they don't provide reproducible builds), it is very much possible for you to know about it, as everything is on your device. Even if the binaries are different from the source code, decompilers, analyzing network traffic, etc. gives the community a good chance at catching malicious updates. Mastodon admins can simply pull up your plaintext DMs on their servers and no one will ever know.

pr337h4m··on Bluesky announces data federation for self hosters
Mastodon DMs have absolutely no privacy: https://github.com/mastodon/mastodon/issues/18079

For a decentralized protocol doing things right is much more important than doing things fast, it is very difficult (and in a lot of cases impossible) to break backwards compatibility.

pr337h4m··on ChatGPT went berserk
This sorta feels like some sort of mathematical or variable assignment bug somewhere in the stack - maybe an off-by-one (or more) during tokenization or softmax? (Or someone made an accidental change to the model's temperature parameter.)

Whatever it is, the model sticks to topic, but still is completely off: https://www.reddit.com/r/ChatGPT/comments/1avyp21/this_felt_... (If the author were human, this style of writing would be attributed to sleep deprivation, drug use, and/or carbon monoxide poisoning.)

pr337h4m··on British intelligence able to read and flag private Snapchat messages
WhatsApp and Signal have had end-to-end encryption for group chats for a very long time.
pr337h4m··on What happened with the Web Monetization API?
If you try to be feeless you become super vulnerable to spam: https://www.coindesk.com/tech/2021/03/11/nanos-network-flood...

IIRC this spam attack cost the attacker only $5000 but rendered the network pretty much useless for hours

pr337h4m··on Building an early warning system for LLM-aided biological threat creation
"However, the obtained effect sizes were not large enough to be statistically significant, and our study highlighted the need for more research around what performance thresholds indicate a meaningful increase in risk."

"We also discuss the limitations of statistical significance as an effective method of measuring model risk"

Seriously?

pr337h4m··on 165 m private submersible superyacht
How are you supposed to hijack a submarine when it's subsurface?
pr337h4m··on Implementing a ChatGPT-like LLM from scratch, step by step
While pretraining a decent-sized LLM from scratch is not financially feasible for the average person, it is very much feasible for the average YC/VC backed startup (ignoring the fact that it's almost always easier to just use something like Mixtral or LLaMa 2 and fine-tune as necessary).

>Introducing MPT-7B, the first entry in our MosaicML Foundation Series. MPT-7B is a transformer trained from scratch on 1T tokens of text and code. It is open source, available for commercial use, and matches the quality of LLaMA-7B. MPT-7B was trained on the MosaicML platform in 9.5 days with zero human intervention at a cost of ~$200k

https://www.databricks.com/blog/mpt-7b

pr337h4m··on Passenger could face $120k bill after F-18s scrambled for Snapchat bomb joke
"An airline passenger who prompted the Spanish air force to scramble fighter jets after he said he was going to blow up the plane he was on appeared Monday in court, the BBC reported.

Aditya Verma was 18 when he and his friends traveled with easyJet from London Gatwick Airport to the Spanish island of Menorca in July 2022.

The BBC reported that before departing, he told a friend on Snapchat: "On my way to blow up the plane (I'm a member of the Taliban)."

Security services saw the message and flagged it to Spanish authorities, who sent two F-18 jets to follow the airliner until it landed, per the BBC."

If the reporting is accurate (a big if, to be fair), this doesn't seem possible unless either the friend reported his message or Snapchat was scanning private messages.

pr337h4m··on CLI user experience case study
Discord has pretty much figured out how to give CLIs an actual UX with their bots and slash commands.

Midjourney is the largest of the Discord apps that have achieved mainstream usage, despite being CLI-only.

← PreviousPage 4 of 8Next →