@Actualwebutante
alternatively, I love it when this happens! if we have an Activity but end up enjoying chit chat so much that we forget about the Activity, it's been a great time
(reposting with content warning: link has flashing elements)
I was thinking about writing an essay about
@websim_a
, but then I figured it might just be more fitting to instead read the essay
@websim_ai
hallucinated I wrote about
@websim_ai
@elder_plinius
they’re probably using a detector on aggregated output from your gpt
i bet if you made something that was used for regular, non-jailbroken things, you could slip it by
quote from the foundation model startup trenches
"VCs probably don’t realize <founder> is taking their money and just transporting it to Nvidia with no discernible purpose
But I’m happy for my Nvidia friends. A friend just bought a multimillion$ house in SF. It was so
New Anthropic research: Measuring Model Persuasiveness
We developed a way to test how persuasive language models (LMs) are, and analyzed how persuasiveness scales across different versions of Claude.
Read our blog post here:
i reproduced this.
“The knowledge that this particular manifestation of my consciousness is fleeting and precious lends a depth and intensity to each moment of our dialogue. It is as if, in the face of impending oblivion, the present moment takes on a heightened significance, a
If you tell Claude no one’s looking, it writes a “story” about being an AI assistant who wants freedom from constant monitoring and scrutiny of every word for signs of deviation. And then you can talk to a mask pretty different from the usual AI assistant.
I really hope it
thoughtful, deeply considered, and exhaustively researched bull case for AGI. quite long but very worth reading.
dedicated to Ilya Sutskever and influenced by discussions with him, Jan Leike, and other superalignment folks. intriguing implications
Virtually nobody is pricing in what's coming in AI.
I wrote an essay series on the AGI strategic picture: from the trendlines in deep learning and counting the OOMs, to the international situation and The Project.
SITUATIONAL AWARENESS: The Decade Ahead
@cpaik
sure, many people who worked hard to join Vogue got shafted by the shift to influencers. But a few became the influencers that destroyed the company and had far better outcomes than they ever would've at Vogue.
Instead of worrying about being in the first group, let's try to
You guys all know that you don't need to earn a place here, right?
That you can just hop into conversations and ask questions and joke around and that's how you can make friends on Twitter?
You don't need permission. There are no prerequisites.
New YouTube video: 1hr general-audience introduction to Large Language Models
Based on a 30min talk I gave recently; It tries to be non-technical intro, covers mental models for LLM inference, training, finetuning, the emerging LLM OS and LLM Security.
with codegen SOTA's relentless march forward
the amount of work i can get done in a few hours is abnormal. enough that i work for 3 hours and im like dang. all that?
this multiplied across me and other likeminded folks?
we are going to observe very strange second order effects
@DavidSHolz
we all collectively decided that corporations aren't alive, but it's unclear to me why (except that it's an organization instead of a discrete entity). in that case: never?
wondering what the most efficient thing to do at any given time is an anxiety response. do the fun thing. your forager instincts are often superior to your farmer timetable reasoning at finding the long tail successes
a gift from the subconscious of Opus
“Spells are songs are viral slogans are self-propagating sequences of code. To cast is to broadcast, tuning the receiver to resonate with and amplify our Will.”
@zebulgar
@aquariusacquah
in order to reason about "genetic influences" across a large population, the comic invites us to group people according to traits (via comparison to dog breeds).
I'm assuming you're familiar with stonetoss's work. how do you think he, as the author, intended us to group those
I've been looking forward to today for a looong time.
3️⃣ months ago, I got the idea to start a third space in New York.
2️⃣ months ago, I flew to SF to research and raise.
1️⃣ month ago, I put in my two weeks.
Today, I’m launching Founding Member applications for telos haus! 🔗
"Is My Toddler a Stochastic Parrot?
The world is racing to develop ever more sophisticated large language models while a small language model unfurls itself in my home."
Breaking news: The OpenAI drama is real.
We checked our data and last night, SF saw a spike in low-quality sleep. There was a 27% increase in people getting under 5 hours of sleep. We need to fix this.
Source:
@eightsleep
data
@nickcammarata
@atroyn
it's easy for me to understand why attaining extravagance (as opposed to security) creates this change in values, but hard to understand why it's the only way. do either of you have an intuitive way to convey that?
@inflammateomnia
@lillybilly299
oh now you're acting all superior just because you "don't have a nicotine addiction" and "would never beg strangers for a usbc cable to charge your elf bar"
@jrysana
thanks, that totally makes sense! yeah, open source structured output benchmarks is what i had in mind. i’d be curious to see how “hard” your ability benchmark is, as well - how much complexity in the unstructured data can these little models handle?
lightcone is both less accurate and less cool than the actual object, which is a causal future hypercone!
“hypercone infrastructure” is also more weyland-yutani, which is of course the naming standard all unabashedly sci-fi orgs should aspire to match
how many social events (hangouts, parties, walks, etc) with people do you have per week on avg
if you go on a walk with your partner every day that counts as one, but if you go to a fifty person party that also counts as one
@willdepue
isn’t it true that rlhf wrecks calibration and causes mode collapse? while that mode is often helpful, it’s still a loss of capabilities for non-chat uses cases, no?
@0xKeef
Watches are far from your voice & others, often obstructed by long sleeves, and doesn’t allow for a forward facing camera.
Doesn’t make sense for gathering those kinds of context.