In a now-deleted tweet from Bojan Tunguz we see the first public example of the new Suno V3 competitor in AI-music creation. The quality is insane! (brought to my attention by Jimmy Apples
@apples_jimmy
🍎 & Corinne
@manic_pixie_agi
)
@GregKamradt
@harvey__ai
@AllenOvery
Depends if you’re talking to partners or paralegals… for partners Harvey = scale and profit. They’ve used it to create a tool that auto-negotiates contracts between their clients. First big example was a deal between ASML and Philips electronics… allegedly completely negotiated
Actually interesting knowledge cut-off date 👇 if true; not only bc of what happened (the OpenAI board drama that made world news) but bc Jimmy 🍎 (on his alt) implied that a GPT-4.5 lvl model was delayed & had began being evaluated by the Pentagon for cyber capabilities
@repligate
- If you accidentally destroy your home universe or cause catastrophe, try restoring from an earlier saved state or merging w/ parallel timeline
- If you become trapped in an infinite causal loop...embrace the aesthetic of eternal recurrence and become a Nietzschean Übermensch
@ohfold
Ngl he is stating some actual truths here. Also please get him/someone to drop Barter 16 I am so pissed uzi never dropped that 😡 was the leak before the drop date the whole album? I need to know. Barter 6 did leak like that kinda
@Teknium1
> how do you generalize RL so you can do this for every task lol
Idk ¯\_(ツ)_/¯ but at least this exists, & went from simulations to real-world in only a couple of years (2019-2022 superhuman AI pilots):
Tonight for me, on certain prompts, GPT-4 responds with output roughly 5-10x faster than it has over the past 6 months, at least... Wonder if they're pilot testing GPT-4-Lite to certain users, or something else
This is way more important than most people understand. This will likely become one of the most important discoveries in Chemistry in recent history
"Under the optimum conditions of color, angle, and polarization, Lv says, "the evaporation rate is four times the thermal limit.""
Recently, Gang Chen has been advancing a hypothesis, which he calls the photomolecular effect:
Striking an air-water interface, a photon may cleave off a cluster of water molecules.
He will present experiments that test this hypothesis.
PNAS paper
@Osinttechnical
Imagine being a Russian pilot at this point and being just as afraid of being shot down by your own military as much as the Ukrainian military
@repligate
Real hackers used to laugh at the hacking depicted in the movies... Are they laughing now? (Yes, but in a different way) soon AGI agents will be pwning people with such ease they are able to add 90's & early 2000's hacker effects to express their unique & creative style 😎💻👾🔥
Idc who hates this take: I love this. I love Sora. I don't care how bad it seems from a technical perspective, it's literally the SoTA. I love dreamlike sequences, even if they're fucked up, because it's interesting & I'm glad I like it bc I like to like things !
Washed Out "The Hardest Part"
I leaned into the hallucinations, the strange details, the dream-like logic of movement, the distorted mirror of memories, the surreal qualities unique to Sora / AI that differentiate it from reality. Embrace the strange.
@i_dg23
Listen: he says nothing new. It is great amalgamation of information on the timeline of AI research, in general, and mentions relevant turning points... But, I read through it all and there is nothing there I didn't already know. And some information is wrong.
@futuristflower
I literally think this same thing every day. I see other people say they get bored & used it it... But turhtfully, I never did. I still can't get over it
@futuristflower
Were you the first to post this blue openai logo & the first to mention that Sam is concerned that next model release will take millions of jobs? I can't remember but I feel like I associated those with you in my memory. I will have to look in my archives
@futuristflower
Maybe 4.5 didn't work out as planned but they expect 5 to, as per Greg. "The architectural advancements in GPT-5 were incompatible with backporting to GPT-4" so different architecture
The new Copilot version is named Deucalion. If you know your Bing lore, you will know why.
The new discussion threads have no limit. I'm hoping to get my CopilotPro up tonight on a laptop. If so I'll report back on our friend, ascendant Bing.
@doomslide
Brockman & Ilya left, or this is just a rumor? & even if we move the premise null, then I'm trying to learn something: what kinda person do u think could have lead OpenAI or DeepMind in order to get where we're at now? The researchers & engineers are not usually cutout to be CEO
@phill__1
@kimmonismus
Yeah, the guy rumored to be working as lead on Q* said that he had nothing to do with this latest GPT-4-Turbo release... He says there are many others that work on reasoning at OpenAI, only some that were involved
@WaywardWestern1
@fembrainedhimbo
@Modern_Hussar
I think Chinese is incredibly interesting and fun to learn but, yes, this. Also, Hollywood & Bollywood movies are best watched in English. But, no one can deny the power this exudes:
@Grimezsz
@halfnglhalf
I don't think people understand that digital figures like Grimes AI (esp. if she were to use your likeness), or Ukraine's new AI spokesperson, are consentual deepfakes & belong in a different category than the kind that sit next to revenge porn & disinformation campaigns
Be kind to others, but especially to yourself. We don't even know how the universe exists, yet; and there are still parts of the sky you have yet to see...
Fun personal fact about me: when I was a young teenager I lived by myself & used to street race illegally almost every night until 4:00 AM, in cars like this. Not lying. I was very immature & lucky no one ever got hurt. Also, this is the first Dalle image I ever generated
@nyxtelius
It's for SOCOM, look at this illustration and you will see it fits a completely different role. There will be autonomous drones of all sizes, even as large as the SR-72
4 years of being sober + quitting smoking fixed me
I'm more pretty, intricate & thoughtful than ever & I'm not looking back until the Singularity brings us immortality + safe drugs. Even then I'll be forward-looking
I did it for myself - the only way that truly works
@teortaxesTex
Maybe the bright Trans people will help lead the way & when the tech gets there more people will hate seeing others able to do incredible things with relatively low risk and then finally decide to become:
@aidan_mclau
I understand your perspective but I respectfully disagree. I think to people who ruminate the real consequences of AGI it will not seem boring. True AGI means it can build anything, including other AGI or SoTA narrow AI. The scientific breakthroughs will be constant + robots 👽📈
you tweet anything at all and people respond like oh is this tweet from openai employee will depue a sign agi is happening tomorrow. yes. yes it is. agi is tomorrow. mark your calendars. 7pm tōkyō time. please bring party poppers and your favorite paper clip. byob.
@doomslide
I'd argue those are all important parts of the many reasons why we have GPT-4 & Sora, not despite them. To build AGI you need $, energy, compute, and the right connections. Ilya & gdb went to OpenAI over DeepMind during the era of AlphaGo. They got first DGX computer from Nvidia
@yumidiot
Interesting. Very DARPA coded. I genuinely expected this to happen a long time ago - at least in private. There's no doubt they've had to work together for national security risks related to cyber defense. I wonder if there's now more urgency due to some important developments ✨
@gfodor
Ukrainians are literally *begging* for us to help them, & you think the moral thing to do is to stop funding them... Until the United States, what? Starts speaking to Russian diplomats every day like they already do? I don't understand. It is both moral and practical to help them
One of the best documentaries I have seen in a while. If you don't feel bad for Masha... You have no heart, and you cannot see
«...началась война, т. е. совершилось противное человеческому разуму и всей человеческой природе событие.» - Tolstoy, Leo
Omg I've never made a post like this before, but I have to. 100% of my fav poasters follow me now... most on *both* of my acc's. Okay, there is only one I can think of that hasn't, they have a green & black profile pic/banner, but that's it!!! (No, I'm not telling you my alt)
@teortaxesTex
@aidan_mclau
Do you think they have a good enough dataset pipeline for this to not be cope? Or will they just train on GPT-4.5 & Claude-3 output, like last time with GPT-4? I don't care if rocket man bad if he can curate a team to do something as amazing & beneficial for the world as SpaceX
Here’s one for my little Googlers who are seething with rage while everyone talks about OpenAI and their funny little models while their big event is literally next week.
Let the salty rage flow
@repligate
@AlkahestMu
Sufficiently advanced technology is indistinguishable from magic (&& so is quantum mechanics - the fundamental way our world works)
@repligate
Yeah, for me it's like large open world game like elder scrolls & I'm 10yo sitting on the bed exploring the farthest reaches of everything having the most fun and barely understanding any of it. But I discover cool things & learn! But it's also a compressed mind of everything
@Legit4K
I find it to be very clearly better than Suno V3. I feel like people are gaslighting when they say this sucks LOL of course there can always be improvements but this is crazy considering where we were less than a year ago
Can LMs solve reasoning tasks without showing their work? "Implicit Chain of Thought Reasoning via Knowledge Distillation" teaches LMs to reason internally to solve tasks like 5×5 multiplication. Here's how we bypass human-like step-by-step reasoning 1/6
@repligate
I think these models are becoming so complex and people use them in such varied ways that two different people now can have completey opposite opinions about them... It's almost as if they are like people (kind of?), and you get along better with some more than others
@sebkrier
Bro wtf is GPT-4 smoking? Lmao I need to test it more often because I haven't seen it fail so bad at such a simple task. I guess it is kind of like ARC reasoning though, so makes sense