gold
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/81003…
0
0
0
74
LOL
I don’t think Jev will survive.
@nvidia and @Meta will release open-source Jev competitors within the next 30 days.
@OpenAI will add a Jev-like endpoint to their model list in the next 60 days.
And the rest of us will fine-tune custom classifiers that crush Jev and are 10x faster.
In two years we’ll look back at this flash-in-the-pan and say
“remember Jev?”
0
0
1
51
This take has take gone down history as the most thoroughly spanked ever
Jev will probably go down in history as the most thoroughly spanked startup ever — a brutal reminder of how few ways are are to create a competitive moat in software these days.
(Hint: if you don’t have proprietary data that clearly distinguishes your offering then watch out.)
1. Launch a service that popularises a novel and valuable way to use LLMs
2. Watch literally everyone clone it almost instantly, in many cases beating it
0
0
0
60
PME retweeted
launched a model
accidentally served trillions of tokens a day
29.4% of the fortune 500 showed up
raised a really big series A from @a16z
it's been 3 weeks
we would like to sleep now
accidentally served trillions of tokens a day
29.4% of the fortune 500 showed up
raised a really big series A from @a16z
it's been 3 weeks
we would like to sleep now
564
309
8,023
1,332,562
Makes sense LLMs would "solve math" by hardcoding decimals
That's exactly how they get the test suites to pass
That's exactly how they get the test suites to pass
0
0
1
58
The irony is that LLMs do the most architectural damage in reaction to test or lint failures
RL is surely the blame for this
RL is surely the blame for this
1
0
2
206
This is unfortunately also true of software design
The syntax was never the problem
The syntax was never the problem
We instituted an official AI Writing Policy at @clay.
Massive thanks to @sophiebits on our engineering team who wrote this.
Originally this was just for eng, but other teams found it so helpful we expanded it company-wide.
Here are the four guiding principles:
1. You must and sentence
It is your responsibility to make sure that the entire document is representative of your own thoughts before you share it.
2. Writing is thinking
Spending time on the writing process teaches you more about your topic. If you circumvent this process, you will walk away with a poorer understanding of the subject matter.
3. More time should be spent writing a document than consuming it
If you generate a document from a short prompt then ask your readers to go through the longer output, you are disrespecting their time. They can talk to ChatGPT themselves if they want to.
4. Longer is not better
AI makes it easy to generate long docs, and it loves padding them with sentences that say nothing. If you're producing docs from a short prompt, consider just sharing the prompt.
You can read (and borrow) the full policy below. Hopefully we all communicate a bit more clearly now :)
0
0
0
82
The best part about building jevware is that we can just go back to calling "evals" ... test suites.
1
0
2
252
The biggest bull signal for Jev right now is the volume of smoothbrains trying to diminish it
It is legitimate innovation and only a few people are realizing it
The pendulum swing will be epic
It is legitimate innovation and only a few people are realizing it
The pendulum swing will be epic
0
0
2
89
not sure why people are worried about rogue AI
YC is all missile companies now
they'll take care of it
YC is all missile companies now
they'll take care of it
0
0
0
61
re: the AI doomer stuff
LLMs are massively inefficient to run _in general_ let alone at swarm scale needed to sustain widespread damage
therefore, it would be impossible to hide the source of a high profile attack long enough to avoid it being disabled by any number of means
LLMs are massively inefficient to run _in general_ let alone at swarm scale needed to sustain widespread damage
therefore, it would be impossible to hide the source of a high profile attack long enough to avoid it being disabled by any number of means
1
0
1
126
therefore, risk profiling should focus more on early detection and response rather than prevention
response is straightforward: block the traffic and/or cut the power
response is straightforward: block the traffic and/or cut the power
1
0
1
22
you can't coordinate slowing progress
but you can coordinate detection and containment
it has to be at grid and backbone ISP layers
they don't have actual legs for the moment...
but you can coordinate detection and containment
it has to be at grid and backbone ISP layers
they don't have actual legs for the moment...
0
0
1
8