ʟᴇɢɪᴛ
@legit_rumors
Followers
6K
Following
24K
Media
639
Statuses
2K
𝒕𝒉𝒆𝒚 𝒂𝒓𝒆 𝒍𝒊𝒌𝒆 𝒈𝒉𝒐𝒔𝒕𝒔 𝒊𝒏 𝒕𝒉𝒆 𝒎𝒂𝒄𝒉𝒊𝒏𝒆 . . .
⍨⧍⍨
Joined October 2018
is the world ready for personal AI agents navigating the Internet? 🧐. language models are vulnerable to A LOT of prompt injection attacks. but I still think best course of action is to deploy them out in the open beyond waitlists, and iterate fast. 1/n
5
3
55
this is my mini benchmark for AGI. if it can see and name them, it's agi. currently no model can see them 👀
342
278
9K
this is so unbelievable true. I’m incredibly grateful that OpenAI shipped gpt-3. the singularity is within reach 😌
36
72
1K
@Willrandship I do not as it changes the image still. but that's a clever way to work around this, nicely done.
4
0
552
you can prompt it how ever you like as long as it's reasonable. I simply hope to see more work put into visual reasoning for 2025.
8
2
523
an updated 2.0 flash thinking model snapshot is releasing in a few days. I do have the date but I'll let them announce it on their own terms 🙂
17
43
442
A conversation with OpenAI on what to expect in the next 12 months. - today's systems are laughably bad.- ChatGPT not long term engagement model.- models capable of complex "work".- like a great team mate working with u.- shift towards verbal interfaces & beyond, Multimodality
28
80
434
here's an exclusive new udio AI music generation 🫡. source: anon. anon is always the source.
47
68
406
Grok 3 launching very soon. I managed to sneak my way in, send prompts fast before i get cut off!
16
17
403
this is so true, and I'm here for it 🍿. every AI lab is cooking, closed & open
7
19
329
the o1 party is over 😢
14
6
299
we're getting multiple Gemini 2.0 models. some are already up for testing on LM Arena including 2.0 Flash (stable) + 2.0 Pro (unclear if it's exp or stable). and there's another thinking model in testing that thinks longer than the latest one to produce better results 👀.
12
14
300
preparations for OpenAI o1 confirmed. the model we got a glimpse of was o1 ✨
6
23
296
AI Studio is ready for the launch of the new Gemini 2.0 models. "Gemini 2" category was added yesterday in the model selector so the stable versions will be listed there as soon as they release. the question is WHEN? and to that idk.
8
20
298
GPT-5 has been spotted by @Jaicraft39 on the official copilot pricing page from Microsoft 👀. surely this can't be a typo right. cc @apples_jimmy @sama
26
39
272
it's time 😮💨. - Gemini 2.0 Pro Exp (experimental).- Gemini 2.0 Flash (stable). we should also be getting a tech report with it!
we're getting multiple Gemini 2.0 models. some are already up for testing on LM Arena including 2.0 Flash (stable) + 2.0 Pro (unclear if it's exp or stable). and there's another thinking model in testing that thinks longer than the latest one to produce better results 👀.
5
10
270
grandma is trapped in this cage until. GPT-5 | Gemini 2 | Grok 3 | Llama 4 😇
15
17
258
oai will make an announcement later today for the release of the o1 model 👽. info provided by @BadArtist154
6
22
263
in collaboration with Kol (@koltregaskes) we've launched a Notion page to track upcoming model updates. ranging from GPT-5 (Orion) & o3, to others like Gemini 2.0 Pro, Claude 3.5 Opus, Claude 4, chatbot arena models
10
11
243
new gpt-4o model updated with June 2024 knowledge cutoff is back 👀. it might be on rollout again so not everyone will have it but go check.
12
7
239
Microsoft has a purchase agreement with Helion for electricity through commercially viable fusion energy by 2028 ⚡. would make alot of sense for the Stargate ai supercomputer for openai to be powered by this as it's also expected to come online as soon as year 2028
Singularity is nearest? .MICROSOFT AND OPENAI PLOT $100 BILLION STARGATE AI SUPERCOMPUTER .MICROSOFT EXECUTIVES ARE LOOKING TO LAUNCH STARGATE AS SOON AS 2028.OPENAI’S NEXT MAJOR AI UPGRADE IS EXPECTED TO LAND BY EARLY NEXT YEAR
16
34
230
@Oli82817545 a theoretical agi would be well informed of pareidolia and be able to identify to a high degree if the image arrangement is an example of such. a general pattern I see as models scale in size is that they become more perceptive but not at the same rate for image modality.
5
1
237
gemini 2.0 flash now available on cursor pro. add it to models list and disable your own api key
9
16
228
anonymous-chatbot is back in the arena. this is usually reserved for GPT-4o model updates inside ChatGPT 👀
5
11
213
@K3vn_C understandable but modern llms should know about Family Guy and approximate cultural aspects. GPT models can sometimes fall for the same illusions as humans so this gives me hope it is achievable. natively multimodal models need more work as we increase reliance on them.
5
0
203
for the record, this is the best news I've heard since the O series of models!. if GPT-5 is on the way, then we are almost certainly going to get another Claude Opus + more from others 👀. even 1 of these alone could be a godsend so I pray that everyone involved can pull it off
@kimmonismus @flowersslop 1 and 2, figuring that out but i think you'll be happy. 3, i would love for us to be able to merge the GPT series and the o series in 2025! let's see.
6
11
197
g(o)(o)d m(o)rning 🌅
good morning.
7
7
197
@JaimeOrtega a pretty useful thing to learn in the context of theory of mind + if we want truly superhuman perception.
1
1
191
calling all ai leakers - blink twice if Llama 3 is still dropping today
14
16
177
- 3.5 Sonnet was not distilled from 3.5 Opus. - 3.5 Sonnet was trained 9-12 months ago. now go read - valuable insight on DeepSeek V3, R1, export controls
My thoughts on China, export controls and two possible futures
5
4
180
A list of upcoming AI model releases. - new gpt-4o / gpt-4.5 both teased in ChatGPT.- grok 3 - in 3/4 weeks from xAI, frontier model.- "pegasus, goblin, gremlin" - gemini 2.0 experiments.- "centaur" - likely gemini 2.0 pro with reasoning.- "maxwell" - unknown, low rate limits 🤔.
6
12
179
Everyone wake up, 2 new OpenAI models in the wild. carrying out tests below, I encourage others to share their results too. let's see if there's any sparks of GPT-4.5 or 4 Lite 🤷♂️
9
10
167
OpenAI Custom Voice Engine ~ found a form to apply for the voice engine api, also found fragments of the code that'll be used to showcase the demo in an upcoming blog post. my understanding is that voice engine is capable of much more realistic and natural sounding voices 🧐
10
12
169
3 things that humanity may never learn. - do aliens exist + where when how.- the true nature of consciousness.- when Google will ship Gemini 2.0 models.
11
7
172
watch OpenAI Sora interact and make changes to the world it creates. "first person POV, Minecraft survival mode"
5
9
168
the entry has now changed to "gemini-2.0-exp". interestingly, there is "gemini-2.0" without exp as well. seems to me that we are getting the stable version of Gemini 2.0 Pro instead of experimental 🤔
6
7
156
New pegasus model in chatbot arena. - identifies as Gemini by google.- has search enabled
6
8
151
OpenAI 12 days returns tomorrow ☃️. I'm most excited for these soon:. Sora | AVM with Vision | GPT-4.5
10
9
144
what did openai research teams see 🧐
something very good has happened.
5
8
144
Gemini 1.5 Flash is going to have significant price cuts on August 12th in Vertex AI. Input cost: $0.0375 per 1M characters.Output cost: $0.15 per 1M characters. HALF THOSE PRICES FOR < 128K CHARS
8
17
134
new updated Gemini 1.5 models might be ready to ship soon™ 🚀. good chance for more than just this; I also expect several new api features 🙂. ETA: 𝐈 𝐝𝐨 𝐧𝐨𝐭 𝐤𝐧𝐨𝐰, my guess would be this week OR early october
5
9
131
@ghstxbt Grok unfortunately picks up on the context of the thread including the answer + hints in some of my replies. you'll need to ask it directly with the image without using the post - it seems to just guess and rotate between a few answers
5
0
131
sora api is live again (will need key ofc) but it no longer redirects you to the front end sora page 👀. this could mean literally anything so don't look too much into it but good to know things are moving again ✨.
7
9
128
o3, like o1, has a little brother. and they're both being prepared! 👀.
6
3
126
New Gemini 2.0 Flash Thinking January 21st version
@legit_rumors Also the new Gemini flash thinking is available on Google AI's Studio.
5
3
125
OpenAI might release o3 on Friday! 🧐.
At 7:30 mark “ o3 which comes out on Friday “ - Chris Lehane Global Policy, OpenAI. Good conversation overall from NPR.
8
1
127
OpenAI Sora Video Generation Model will be coming to Adobe Premiere Pro later this year 👀
5
7
119
gpt2-chatbot speculation. "model𝘀" - Jason (@ OpenAI). "release𝘀" - lmsysorg. 😶 😮 🥲
3
1
113
DeepSeek R1 now has a HF Repo 👀. This is their full, big reasoning model like o1 so I can't wait to start using it!. start observing the repo and the DeepSeek API for a potential release.
New @deepseek_ai . R1 . Obligatory @teortaxesTex . Empty for now.
2
2
119
Gemini 2.0 Flash (stable) ✅. Gemini 2.0 Pro *might* be in “next few weeks” since they are removing 1.5 Pro?. strangely, the API does not have this update yet. Reminder that you promised us a GA version in January - you no longer have days / weeks. Ship it!
📣 Gemini is now powered by 2.0 Flash, our latest model that is designed for the agentic era, delivering fast responses and stronger performance across a number of key benchmarks for everyday help with tasks like brainstorming, learning, or writing. Rolling out to all users
6
5
118
I've been given a time frame and it's a bit unusual. consider it to be "too good to be true" but if you want to join along, it's going to be a couple hours from now. .
oai will make an announcement later today for the release of the o1 model 👽. info provided by @BadArtist154
9
7
113
Say hello to gemini-exp-1121! Our latest experimental gemini model, with:. - significant gains on coding performance.- stronger reasoning capabilities.- improved visual understanding. Available on Google AI Studio and the Gemini API right now:
2
6
112
A new GPT-4 Competitor has risen 🎊. Reka Core Unpacked 📭. 🚀 Generally Available now!.🧠 MMLU 83.2% vs 86.4% GPT-4.💪 Comparable to GPT-4V.🏃 Further training in process!. "Outperforms Claude 3 Opus on Multimodal human evaluations and surpasses Gemini Ultra on video tasks" 🤔
Meet Reka Core, our best and most capable multimodal language model yet. 🔮. It’s been a busy few months training this model and we are glad to finally ship it! 💪. Core has a lot of capabilities, and one of them is understanding video --- let’s see what Core thinks of the 3 body
8
22
110
OpenAI updated their pricing page to show as "per 1M tokens" by default. incoming long context gpt model? 👀. Also, free use of retrieval in Assistants API has been extended until April 1st - thank you 🫶
11
7
108
Grok 2.5 - "our most intelligent model". Along with its own site + new logo? 👀
Grok[.]com is possibly coming soon with Grok 2.5 model (grok-2-latest - "Our most intelligent model") - thanks for the hint, anon!
2
3
112
Interesting. 🤔
GPT-4o updated. Knowledge cutoff is June 2024.
4
4
112
"we plan to launch o3-mini around the end of January with 03 following shortly after". - samta.
2
2
110
gemini 2.0 is omnimodal? swish has found that it supports image and audio outputs 👀.
"Gemini now supports image and audio generation. To try these features, select the gemini-2.0-flag-exp model. You will then be able to change the response type to "audio" or "images and text" to try it out.".
2
3
108
sam: ❝ very good releases coming later this year ❞. no model named GPT-5 tho
5
7
108
Grok 3 pretraining is now complete 👀. I assume post training is next; molding its character, response style, etc. I've been impressed with how well integrated Grok has become on X.
Cool!. And Grok 3 is coming soon. Pretraining is now complete with 10X more compute than Grok 2.
3
4
105
@MicahJanke not sure if this will help but try telling it to maximize reasoning effort / use more reasoning tokens. I'm curious if it can eventually solve it.
1
2
101
Gemini Update delayed to June 18th
2
5
98
I wonder what could've been if we got gpt-4.5 / gpt-5 instead 🤔. how much time could it free up, how reliable could it have been, how much better could they make it, etc. I miss the GPT series already
11
1
101
just a reminder - Lex Fridman will host a new podcast ep with Sam Altman "soon". he also did this last year shortly after gpt-4 was announced 🙂
5
7
94
@sithamet haha, it's Peter Griffin & Brian Griffin from Family Guy. a theoretical AGI should be able to perceive that the user is trying to induce pareidolia in them to "see" those characters.
5
1
99
anonymous-chatbot is still in the arena, with the June 2024 knowledge cutoff previously teased in ChatGPT. it's been in the arena for a few weeks now, but I will continue to monitor this + a few other upcoming models from various labs 👀
3
2
97
a new upcoming model from anthropic. "anthropic-firefly-preview". it could be a new entry in the Claude family of models OR an entirely new series of models from Anthropic 🧐.
BREAKING? 🚨: What do Anthropic and Firefly have in common? 👀. But aside from this, I am impressed with how many AI models is Notion integrated with 🤯
4
5
96
udio v1.5 model dropped yesterday. here's a comparison - (left: v1 right: v1.5)
3
9
94
something's happening at OpenAI 👀. what do you think it could be?
3
9
92
gemini 2.0 flash party is over 👋
gemini 2.0 flash now available on cursor pro. add it to models list and disable your own api key
2
6
94
@sama tip: don't mix the two series like gpt-o5. if the merge is successful, I'd recommend sticking with the O series naming but without the low, high etc . put those behind a config instead and continue with: o4-mini, o4, o4-pro. but 4o needs to go when/if o4 drops ❗.
12
0
96
the bottom line from these early Gemini 2.0 Pro experimental versions is that post training can yield sizable gains. look at the difference between the different exp-date model snapshots 👀
2
2
94
gemini web model selector updated just now. new layout accomodates for multiple models to be listed under each tier 👀
7
4
92
Grok 3 to be released in 3-4 weeks. FYI - this is measured in elon time ❗. if true, that's around the time of o3-mini
Grok 3 is "3-4 weeks" away. No source, also this is Elon-time. ;-).
6
0
92
@chefronadin "the concept of pareidolia, mentioned in the thread". Grok is reading the replies to retrieve the answer, a useful feature in general but counterintuitive for the test.
1
0
90
Who did it better? Give your reasoning. Gemini 1.5 Flash or ChatGPT (4o)
16
3
87