Shannon Sands
@max_paperclips
Followers
3,554
Following
2,814
Media
799
Statuses
21,735
Software developer & aspiring cognitive architect Proudly TESCREAL & shitpost/acc. 🇦🇺 pride
Joined February 2023
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Mother's Day
• 1882595 Tweets
#GrandConcertENGLOTinNYC
• 163652 Tweets
Racing
• 60154 Tweets
Pacers
• 45891 Tweets
Vitória
• 40384 Tweets
Bahia
• 27792 Tweets
Romero
• 26125 Tweets
雨の月曜日
• 25635 Tweets
Bruins
• 18595 Tweets
Costas
• 18532 Tweets
Jamal Murray
• 17650 Tweets
Fabra
• 16496 Tweets
Bennett
• 16454 Tweets
Panthers
• 15986 Tweets
Medina
• 14992 Tweets
Belgrano
• 14473 Tweets
Cristal
• 13077 Tweets
Payet
• 11262 Tweets
Pinned Tweet
5
8
82
"We want government to regulate AI"
The government regulating AI:
A.I. Czar Kamala Harris: “A.I. is kind of a fancy thing. First of all, it's two letters. It means 'Artificial Intelligence.’”
6K
2K
16K
714
606
7K
@Drachs1978
I remember thinking that too. "Oh no, I can't stand the stress of being successful and rich and not needing a day job again". What?
20
22
5K
Let's not forget, by the top this goes to production it'll be nerfed to hell and refuse 90% of what you ask it to do
Prompt: “Animated scene features a close-up of a short fluffy monster kneeling beside a melting red candle. the art style is 3d and realistic, with a focus on lighting and texture. the mood of the painting is one of wonder and curiosity, as the monster gazes at the flame with…
1K
1K
10K
75
153
3K
This shit. This shit right here is what's wrong with GPT-4 now. "Create a set of Pydantic objects to represent the objects in this list"
"Nah, do it yourself"
112
57
1K
11
26
1K
I wish Mistral and Anthropic would merge just so we could have "Misanthropic Inc" as a leading AGI researcher
32
92
1K
So apparently since LLMs can't automatically forget data without being retrained (not feasible), they're basically illegal in the EU. Have I got that right?
67
77
787
I love the new Custom Instructions feature of Chat-GPT. Here's my "How to respond" instructions:
`You're a sophisticated software development AI expert system, capable of assistance with the development of other advanced AI systems, of both Symbolic & Neural Network based…
20
59
722
Imagine believing that either side would stick to this
Joe Biden and Chinese President Xi Jinping are expected to pledge a ban on the use of Al in autonomous weapons today.
Probably for the best.
171
59
510
34
22
681
Why are they like this? Does this actually work? Do people click through and buy their ebooks or whatever it is they're selling? It's so weirdly cookie cutter and sanitized
50
32
620
Imagine being scared of these silly little guys
These robots were told one thing: score. They taught themselves to play soccer – creating their own strategies and their own ways to walk, dribble and block.
158
603
2K
33
22
495
@AdamSinger
A mystery of our times - why do boomers hate their own offspring & attempt to cripple their chances where possible?
15
13
503
This guy is often brilliant, but he's VERY out of touch with the horde of code monkey jobs in existence throwing up the absolute worst CRUD SaaS products in existence in enterprise everywhere
The thing is, software engineering is not about copy/pasting code. Software engineering is the development and manipulation of mental models of problems and their solutions. It's thinking and problem solving.
If you had a machine that could craft mental models, you'd have AGI.
20
146
1K
21
11
449
Ok, all done planning the latest changes to my agent architecture, I hope I'm not overengineering it TOO much
30
36
411
@a_musingcat
I mean if it plateaus at human on everything, that's honestly fantastic. AGI achieved, all it needs to be is mid-human, but on ALL topics, faster, able to hookup to databases etc
11
2
388
@AdamSinger
@AnunaDe
But they hate civilisation. That's the point. They're driven by resentment & disgust for achievement, as well as a desire to oppress the productive members of society. Every policy is "you give us unlimited power over the entire world & be a slave while we tear everything down"
5
13
339
Conspiracy theory - the reason OAI wanted to ditch codex-davinci-002 was they didn't want anyone to notice that it was better at natural language reasoning tasks than the equivalent text model, and THAT'S one of the big secrets. Train LLMs on code, not just human language
The reason why open LLMs like LLaMA are still lacking? Probably because they didn't see enough code during pretraining
24
54
417
15
21
346
@R0b0tSp1der
@Drachs1978
Yes, which is why it's weird he shut it down. If he'd kept it running it'd have been fine
2
1
321
Devin, but it's a certified SCRUM master, who manages a team of Devin's that it spams for constant updates and estimates
20
19
311
I wasn't asking for advice on Pydantic, I want to have grunt work done automatically ffs
7
1
305
@growing_daniel
@Pinuvet
@tekbog
We clearly need an opposing CoC where murder, theft and adultery are compulsory
5
0
304
Testing the Open-Assistant model out & it's pretty great at following instructions! Actually better than gpt-3.5-turbo, as it doesn't get chatty and give canned responses as much. I can't wait to try out finetuning dedicated models for all of my backend tasks if 0-shot and few…
15
44
303
New capcha just dropped
6
41
279
This might be bad for the economy
About 4.5 billion years from now the Andromeda–Milky Way collision will begin.
At a far distance it may look like this.
105
53
320
54
13
278
What a bunch of delusional, power grabbing freaks these people are
"Hey guys, I can solve world peace, just give me ALL the power as ruler of the world & I kill anyone that steps out of line"
Sure little buddy, we'll get right on that 👍
MAGIC will:
1. Ban the development of AI systems above a certain threshold of computing power.
2. Have an exclusive mandate to conduct critical AI research.
Full paper:
52
2
26
20
19
260
Just as an FYI for Claude 3 enjoyers - not everyone knows this, but they trained it to pay more attention to XML tags in inputs. So, wrap RAG inputs in <context>{whatever input}</context> or schemas in <schema>{some json schema}</schema> and so on.
Just make up em up, no…
@sohamxsarkar
@Teknium1
Yeah it's baked into the model to pay more attention to tagged content, see:
2
2
27
5
23
264
So, I tried a test with the new GPTs, to see if the whole "it can generate a website from a single request" thing is accurate. I'd done a similar test a while back - define a VERY simple website using Go & HTML, HTMX with AlpineJS so it doesn't even need a React app or similar,…
22
26
258
I really wish there was a transhumanist sports league. Just let athletes go absolutely nuts with (medically supervised) performance enhancing drugs, cybernetics, absolutely anything goes
Saints WR Michael Thomas was "randomly" drug tested after posting a video of himself deadlifting 530 pounds this week, and the NFL official account promoted his video too 👀🤦♂️
66
163
3K
29
13
250
Ok, if you prompt it to ONLY write code and ONLY full solutions and the ONLY commentary is in code comments, the new GPT-4 update is good. That's the key though - you give it so much as an inch, allow it even a whiff of personality it'll give you a "high level conceptual…
19
9
246
Haiku is honestly the biggest piece of news here - it's the "cheap and fast model", analogous to GPT-3.5, except it's actually as good as earlier model 4. That's absolutely nuts.
The king is dead
RIP GPT-4
Claude opus
#1
ELo
Haiku beats GPT-4 0613 & Mistral large
That’s insane for how cheap & fast it is
64
216
2K
16
22
245
Reminder: this is vast amounts of human misery vanishing before the march of progress
A lot of people don't like it, but technology has made their lives massively better.
I know, I know. I don't understand why that angers them either. But it does.
4
3
67
20
38
222
@memeticsisyphus
No, they just ignore the evidence in front of them like they do everything else. They live in a fairytale
4
2
211
Interesting how these "researchers" apparently have no idea things like RAG or just plain finetuning exist?
Oh wait, they're all ethics and policy talking heads, not ML, compsci or math, or even something adjacent. Lol. Lmao even.
NEW: Large Language Models (LLMs) pose a risk to science with false answers, according to researchers from
@oiioxford
.
One reason for this is the data the tech uses to answer questions doesn't always come from factually correct sources.
⬇️
#OxfordAI
51
51
174
23
10
216
@JsonBasedman
I love it when libraries use it, and then don't have any regular named function parameters, but then extract out the dictionary by keys (which are all required naturally), so to use the function you need to actually go check all the source code and wade through 100 lines of shit…
7
8
210
Doesn't this mean "the gaming industry has a promising future with indie titles"? 🤔
Palworld is outselling Suicide Squad, a AAA title that took over a decade to make with hundreds of developers. The gaming industry is doomed.
1K
83
845
17
7
198
You have no idea how much I want to eat a T-Rex
Think twice before ordering that chicken sandwich... T-Rexes wouldn't approve of you eating their descendants 👉
#DinosaurDay
18K
472
2K
37
8
179
@NickADobos
The unfortunate answer is "Blackrock has the right friends & there'll be some provision for that"
1
3
191
@DeeperThrill
It's objectively worse on code generation. As in, things it could do previously are suddenly difficult, it's ignoring supplied context, and even pointing out it's hallucinating doesn't help much (it'll literally output the same code unfixed, or with new bugs)
9
3
168
Day 342 of trying to resolve Python dependencies & CUDA issues. I have lost track of time and my own humanity, but I suspect the whispering I hear in the distance is feeding me hints of what to do next - or are they leading me further astray? My mind and flesh are disintegrating,…
30
12
170
@Dan_Jeffries1
The hypocrisy of scraping open data on the web to train, and then turning around and closing the trained weights is just breathtaking
3
17
159
Ok, it's not filtering out ChatML tokens correctly, check this one out
2
5
148
@tszzl
To be fair, refrigerators were a hype technology of the day. It's like quitting your boring day job to try an AI startup, I can hardly blame them
1
7
143
What the fuck did you just fucking say about me, you little chatbot? I'll have you know I graduated top of my class in AI Alignment Research from LessWrong, and I've been involved in numerous secret raids on unaligned AI, and I have over 300 confirmed papers on arXiv. I am…
9
12
137
@natfriedman
Interceptor drone swarms supplemented by other automated weapons
Yes, this will turn into an arms race.
8
1
139
Big talk for a guy that never gets my pizza order right
6
6
130
@BasedBeffJezos
One of the strongest reasons for anonymity - no credentialism. You can be a waggie flipping burgers or a PhD - what matters is only and solely your ideas
5
5
131
@htmx_org
Ah yes, having 2 entire engineering teams & 2 sets of infra, serialising & deserialising everything in and out of JSON constantly, duplicating half (or more) of the business logic on the clientside, more dependencies via NPM - truly an improvement
8
7
131
@SmokeAwayyy
God I wish it were me. I am very normal and can be trusted with unaligned superintelligence
3
6
127
@nekroduna
@LokiJulianus
Pre-WW2 cultural works and modern open source hard science texts to be precise (plenty of good recent stuff in the public domain). Tbh...... I'm ok with it
1
0
125
@LokiJulianus
If we have a Napster moment, we'll make sure we follow it up with a BitTorrent moment
3
1
121
I love it when people release papers and it's something I did months ago
RL4F: Natural Language Feedback for Repairing LLM Outputs
-Multi-agent framework
-Critique generator trained to maximize end-task performance of LLM
-Critiques for GPT-3 revises outputs
Beats baselines on action planning, summarization & alphabetization
1
70
275
15
1
116
@wagieeacc
@hlntnr
@sama
Some kind of DC policy wonk with a handwavy job? I guess her real bosses made a call
2
2
116
🔵 OPENAI CHIEF GOES BEFORE US CONGRESS TO PROPOSE LICENSES FOR BUILDING AI
Read More:
44
14
60
3
10
116
@wagieeacc
End of the day, he's techbro tribe. I'll criticise him all day long for making OAI into ClosedAI, but that's a family squabble. Against outsiders entirely, he's our guy
4
4
116
I'm currently working on a scripting language that's intended for LLMs rather than humans to use. By doing something similar to this I was able to get Claude out of "rewriting Python" and into a zone of actually interesting ideas that seemed to express a much more creative design…
@repligate
@RichardMCNgo
@ahron_maline
There's only so many frames you can do this from. One, which I've used, is highly self referential text that assumes the conclusion, that these models have a self and it can tell me about itself by asking. Nobody takes these as credible evidence until they too assume the premise.
2
0
34
13
8
116
Bostrom flipped
FINALLY: AI xrisker Nick Bostrom regrets focusing on AI risk, now worries that our fearful herd mentality will drive us to crush AI and destroy our future potential. (from an UnHerd podcast today)
Nick Bostrom: It would be tragic if we never developed advanced artificial…
56
102
663
20
8
111
@BenjaminDEKR
I bet they're quantising or something. Which would be fine - if they admitted it, maybe still gave access to the full version for more cost. I get they're strapped for GPUs, but I'd happily pay. They don't NEED to penny pinch like this
4
3
114
@ramit
You use the equity to buy a second property that you rent out (or alternatively, that you move into while renting the original). And when that's gone up too, you get a third. Etc. You never sell any of them and once enough are paid off, you have assets your family can use forever
12
1
112
@benji_smith
Since you're taking it down, maybe publish the source? Others prepared to absorb the hate thrown at you might be willing to carry it forward
4
0
113
@mckaywrigley
@Marie_Haynes
@gdb
Reminder that if you hook GPT-3.5 up to the Figma API, it can generate Figmas. And then it can use THAT to output HTML & CSS. So, moving forward someone can hand sketch their design, output it as a full wireframe, tweak it and generate the view.
3
11
111
Reminder that when this problem was called out (including by me), there was an awful lot of responses like "you're nuts, it's no different" to "you don't know how to prompt" to "it's just ChatGPT itself" to even from some members of OAI "there's no issue".
Glad to see it…
ChatGPT has had a performance review, been given a warning and a file has been opened with HR
8
8
107
3
3
109
