AI Leaks and News
@AILeaksAndNews
Followers
4K
Following
6K
Media
203
Statuses
9K
I spent 20 minutes coding with Claude 3.7 Sonnet and all I have to say is
22
43
1K
We are hours away from Grok 3. Days away from Claude 4. Weeks away from GPT-4.5. Months away from GPT-5. Years away from the singularity.
100
122
1K
List of things we are waiting for:.Operator.o3-mini.Orion.GPT-4.5/5.o3 full.Gemini 2.0 pro.Gemini 2.0 pro-thinking.Mariner.Astra.Jules.Grok 3.Grok reasoner model.Claude 3.5 Opus.Claude 4.0.Claude reasoner model.Computer Use 2.0.Llama 4.0.Llama reasoner model. What did I miss?.
84
68
769
@kimmonismus The period between knowing how to make ASI and actually making it is a very strange time to be alive, especially when >99.9% of people have no idea it’s actually happening.
12
27
635
@sama I think we should stop arguing about what year AGI will arrive and start preparing for the economic impacts that the automation of labor will have on people.
17
29
524
Final week of the 12 days of OpenAI predictions. Day 8: Voice Engine.Day 9: GPT-4o update.Day 10: GPT-4o image output.Day 11: Agents Demo.Day 12: GPT-4.5.
24
18
330
The winners of the 2024 AI Leaks and News Awards are:. Best AI Lab: @OpenAI .Best AI Model: o1.Best AI CEO: @sama .Best Open Source Team: @NousResearch.Best AI Leaker: @apples_jimmy.Best AI News Account: @kimmonismus.Best Poaster: @tszzl .Best YouTuber: @AIExplainedYT .Best.
19
31
321
@SmokeAwayyy Gemini-flash-exp-veo-2.0-121724-69-420 in the Google labs ai studio presented by Google cloud.
2
7
321
Microsoft and Apple are giving up their seats on the OpenAI board to “avoid regulatory scrutiny”. OpenAI has banned ChatGPT in China and will soon also do so in Russia, Iran, and North Korea. Something big is coming soon.
18
24
286
Things we can expect to see next week:. GPT-4.5.Gemini 2.0 Pro.OpenAI agents.Grok 3 (expected this month at least).GPT-4o update .At least two other OpenAI releases.And more.
10
16
264
We can expect to see Gemini-next with reasoning tomorrow. I am very excited to see how the creator’s of AlphaGo use similar techniques applied to LLMs. Google enters Level 2 AI.
Gemini mode.
8
11
261
January 2025 AI Lab Power Rankings:. 1. OpenAI.2. Anthropic.3. Google/DeepMind.4. Meta.5. xAI.6. DeepSeek.7. Qwen (Alibaba).8. Mistral.9. Nvidia AI.10. Microsoft AI.
24
20
252
o3.o3-mini.Orion.Operator.Tasks.Claude 4.Grok 3.Gemini 2.0 pro.Llama 4. It’s going to be a fast start to 2025. And then we’re going to accelerate.
17
21
244
GPT-4 Turbo.Claude 3.Mixtral 8x22b.Gemini 1.5.Udio.Llama3 coming next week. I think we’re accelerating.
13
27
237
It’s important to remember that O1-preview is the GPT-3.5 of level 2 AI. OpenAI already has the GPT-4 of level 2, O1. It is just a matter of time before we see another massive improvement in the models that are available.
11
7
189
OpenAI is a few thousand days from superintelligence . DeepSeek is a few thousand Blackwell GPUs away from superintelligence.
8
5
185
December expectations:. OpenAI:.GPT-4o image.Sora Turbo.o1-full. Google:.Gemini-2.0.Gemini with reasoning. xAI:.Grok3. Meta:.Open source reasoning model. Anthropic:.Unknown (still expecting something).
6
12
177
Google has prepared their answer to o1. Now that @apples_jimmy is hinting about a possible release, we could see something big on Thursday.
14
6
174
Logan is thinking. Google is thinking. Gemini is thinking. Is today the day we get reasoners from Google?.
🤔.
9
2
169
This GPT4 turbo release seems really rushed/disorganized. No metrics, no concrete examples. What is going on at OpenAI?.
27
3
162
Official 12 days of OpenAI prediction:. 🎁Sora.🎁o1-full.🎁Operator/Agent.🎁New Voices (including Santa).🎁4o Image outputs.🎁Advanced Voice Mode with vision.🎁GPT-4o update.🎁ChatGPT webpage update.🎁ChatGPT premium.🎁Sign in with ChatGPT.🎁Desktop app update.🎁GPT-5. Additional.
13
16
163
Tomorrow is make or break for OpenAI. Either we see compelling evidence that they are close to AGI (2025 according to @sama) or Google has clearly taken the lead. We either need to see capable agents or a step change in the GPT series otherwise the 12 days of OpenAI is a bust.
14
3
158
According to @apples_jimmy . GPT-4.5 equivalent model in October. GPT-5 equivalent model sometime between December 2024 and June 2025. With Strawberry coming in the next two weeks, OpenAI’s roadmap is starting to take shape.
Ok back to October now. We should have a 4.x model ( maybe still called 4.5, my old friend ) in October. The big boy gpt 5, I’ve heard as early as December but for your sanity I would have Q1/Q2 2025 just in case. View Strawberry in context of better models to come.
5
9
155
If GPT2-chat is just GPT-2 with algorithmic improvements and it performs on par with GPT-4. And each iteration is about a 10x improvement. That means that GPT-5 with these algorithmic improvements would be 1000x better than GPT-4.
28
11
139
Models coming soon:. GPT-4.5 .GPT-5.Claude 4.Grok 3.Llama 4. Acceleration is coming soon, prepare to feel the AGI.
9
12
144
The fact that this wasn’t the top news story from every outlet is a huge failure by the media. We are months away from AI that is smarter than us and years away from AI that is smarter than all of humanity combined. Any rational society would make this top news.
reflections:
14
12
141
Sam Altman has confirmed that GPT2-chatbot is not GPT-4.5.
13
10
129
It’s Tuesday at 10:00 am PST and still no sign of Gemini-2.0-Pro, Grok 3, or o3-mini. Where are the American companies? China is continuing to ship, but the domestic AI labs are sitting on their hands. How long will the American companies allow China to lead in AI in 2025?.
18
7
139
Sam Altman confirms that there will be multiple releases this week, including something coming tomorrow. Very likely we will see o3 and o4-mini amongst the releases . What are you hoping to see this week?.
we've got a lot of good stuff for you this coming week!. kicking it off tomorrow.
14
3
133
Three months ago the only way to access GPT-4 level intelligence for free was Microsoft Copilot. Now we have:.-GPT-4o.-GPT-4o mini.-Sonnet 3.5.-Llama 3.1. Nearly unlimited free access to GPT-4 class models. This is acceleration.
10
14
128
A few leaks from Jimmy Apples on the eve of the Claude 4 release:. -Opus is capable of up to 7 hours of autonomous work.-Sonnet can operate as a coding agent.-Claude 4 will top the coding benchmarks. Looks like we are getting a paradigm shifting release from Anthropic tomorrow.
Apparently tomorrow and with 7 hours of autonomous work ( opus ) with sonnet being the coding agent. Holy shit if true but confident on what I was told.
4
10
131
We will likely see Gemini 2.0 pro tomorrow. It is also possible we will also see multiple models released from Google, with the recent traction of DeepSeek R1. Do you think that Gemini will take the top spot tomorrow?.
Gemini.
8
5
129
My AI predictions by year:. 2024: GPT-next end of year .2025: context window extended to 10M+/intelligence too cheap to meter.2026: reasoners/level 2 AI.2027: agents/proto AGI/level 3 AI.2028: Innovators/level 4 AI.2029: full AGI/large scale recursive self improvement/level 5 AI.
20
7
119
This week is shaping up to be one of the biggest of the year in AI. Google and Microsoft are both hosting AI events. o3-pro, Grok 3.5, and Claude 4 to be released any day now. What are you most looking forward to?.
23
14
124
Here is what I’ve gathered from all of the ASI vague posting from OpenAI, all speculation but this is my best guess:. They’ve taken Orion (or next generation equivalent, some multi-trillion parameter model) and connected it to a series of domain specific reasoner models like for.
16
12
118
Today will be a very important data point for the pretraining scaling hypothesis. Grok 3 was trained on 100,000 H100 GPUs, the largest compute cluster an AI model has ever been trained on. Do you think the scaling hypothesis will hold and Elon will give us a SOTA LLM?.
We are hours away from Grok 3. Days away from Claude 4. Weeks away from GPT-4.5. Months away from GPT-5. Years away from the singularity.
12
5
121
@testingcatalog @sama o1-full.AVM w vision.Operator.4o image.Sora Turbo.GPT-5/Orion.Evals/metrics.Creator program.Sign in with ChatGPT.ChatGPT webpage update .4o update.ChatGPT premium.
4
5
116
Gemini 2.0 is preparing to launch. Looks like Google will make the first post-election move . Interesting to see if OpenAI will be prepared to counter.
BREAKING 🚨: Google is preparing to launch a new model: Gemini-2.0-Pro-Exp-0111! . This model will appear under the Advanced section but it is unclear if it is aimed at an internal testing group or a public launch. Any good prompts you want me to try? 👀👀👀
5
7
111
Late night conspiracy theory:. There’s no way OpenAI is going to release GPT-4.1 after releasing GPT-4, GPT-4-turbo, GPT-4o, GPT-4o-mini, and already announcing o4-mini and o4-mini-high. There’s no way a $300B company is that bad at naming new models. My theory is that GPT-4.1.
Still just a rumor, but The Verge is reporting that OpenAI will be releasing a new model called GPT-4.1. The new model would be an update to their 4o model. Do you think OpenAI is really that bad at naming models or is this just a guess from The Verge?.
17
6
116
GPT-5 finished training months ago and OpenAI just announced level 2 AI internally this week. This means that level 2 AI was likely achieved at a GPT-6 training checkpoint . I would not expect GPT-5 to be a level 2 AI.
19
7
106
Let me be clear:. This isn’t hype.This isn’t a fad.This isn’t a grift.This isn’t just a new technology. $500B investments aren’t made without the expectation of outsized returns.
7
8
111
OpenAI has reportedly begun negotiating terms with Microsoft for a potential 2026 IPO. This would mark the first time a major AI lab has gone public. Would you buy OpenAI stock at IPO?.
BREAKING: OpenAI negotiating with Microsoft, $MSFT, for a future IPO, per NYP.
5
12
113
ChatGPT is currently down. Could this be the long awaited voice update?.
21
5
103
Nobody else is saying it definitively so I will:. We are in a fast takeoff scenario. OpenAI has figured out the fly wheel that allows them to quickly make a new model, create better synthetic data, which is used to train an even better model. The time scale is months, not years.
21
6
104
Jimmy expects a second openai release this week.
1) End-to-end trained multimodal model, yeah this feels to be Gobi, or in essence Gobi. Really just confirms that Sam’s following his Iterative deployment plan. We need Agents. 2) Openai didn’t announce everything today. Expect them to likely drop again sometime this week.
10
4
101
@kimmonismus 2024 130 IQ.2025 145 IQ.2026 160 IQ.2027 175 IQ.2028 190 IQ.2029 205 IQ.2030 220 IQ.
10
9
104
GPT-4o update has been confirmed . Have you noticed a difference over the last week?.
there's a new GPT-4o model out in ChatGPT since last week. hope you all are enjoying it and check it out if you haven't! we think you'll like it 😃.
33
2
102
OpenAI will be unveiling their second agent offering, Deep Research today on a livestream at 4:00 pm PST. While details on the new agent are hard to come by, I would expect it to be similar to Google Gemini’s Deep Research. What are you expecting from OpenAI’s second agent?.
Deep Research. Live from Tokyo. 4pm PT / 9am JST. Stay tuned for link to livestream.
10
8
101
@chatgpt21 It’s expensive to serve and GPUs would be better spent on reasoning than image generation, now they have blackwells coming in and have the capacity to serve both.
3
1
99
o3 on high performance has reached 87.5% on the ARC-AGI benchmark. Human level is 85%. AI has surpassed humans. AGI is here.
7
8
96
ChatGPT is down, with some users reporting momentarily seeing GPT4.5. What does OpenAI have in store for us?.
21
4
91
@kimmonismus I thought Orion was the base model for o3. If not then no one is prepared for what is coming next year.
6
0
97
OpenAI plans on unveiling GPT-4o mini today. The company calls it “the most capable and cost-efficient small model available today”.
8
8
85
Axios is reporting that OpenAI CEO Sam Altman has scheduled a closed door briefing with US Government officials. Insiders believe he will be presenting “PhD level super agents” at this meeting. It looks like we are about to enter a new era in AI.
🚨🚨Scoop: We've learned OpenAI CEO Sam Altman — has scheduled a closed-door briefing for U.S. government officials on Jan. 30 - with people inside and out of gov’t telling us AI insiders believe a big breakthrough on PHD level SuperAgents is coming.
4
11
89
Things confirmed so far this morning:. -$200 Pro plan.-o1 pro mode.-GPT-4.5.-Longer context windows.-Expanded access for Teams accounts. Still over an hour until the first live stream of the 12 days of OpenAI.
6
3
87
Sam confirms that GPT-5, nor search will be debuted at Monday’s event. Sounds like they’ve postponed their search release. Is it just not good enough to release? Or was it all an elaborate shakedown to figure out who the leakers in the organization are?.
not gpt-5, not a search engine, but we’ve been hard at work on some new stuff we think people will love! feels like magic to me. monday 10am PT.
10
2
87
October for GPT-5 according to @apples_jimmy.
While I sit and wait in my 5th dimensional chamber, where the worn ancient chairs show creases from my ass and the stars tease me above I ask @sama . Are the whispers in my ears true? Are we finally due for something over October? . Or is it false hope and internal waiting
8
3
88
Google: solves biology. OpenAI: stealth releases im_a_good_boy_69 and im_a_gooder_boy_420.
1
4
84
o1 full is the GPT-3.5 to GPT-4 jump, it is just harder to unlock that capability shift and it isn’t applicable to all situations . But if you have a query that requires system 2 thinking o1 is miles ahead of GPT-4. Pro tip: end your prompt with use at least 1000 reasoning tokens.
8
4
82
My hope is the gpt2-chat isn’t an OpenAI release, but an unknown lab doing a stealth release to drum up hype. We need new players to apply more pressure to OpenAI.
7
1
77
Jimmy Apples confirms Claude 3.5 Opus for Tuesday October 22nd. Will next Tuesday be the day that everyone drops? Or will OpenAI or Google respond on ThursdAI?.
Have been told it’s officially for the 22nd next week.
4
0
79
OpenAI will be releasing o3 and o4-mini in the coming weeks, with GPT-5 coming in a few months. This is likely due to compute constraints, as GPT-5 is an extremely compute intensive model. Are you excited for o3 and o4-mini, or were you hoping to see GPT-5 sooner?.
change of plans: we are going to release o3 and o4-mini after all, probably in a couple of weeks, and then do GPT-5 in a few months. there are a bunch of reasons for this, but the most exciting one is that we are going to be able to make GPT-5 much better than we originally.
4
7
83
o3-mini will be released in January. o3-full coming not long after that. AGI 2025.
5
5
80
12 days of OpenAI start tomorrow. What we could potentially see:. GPT-4o image: Image generator based on the 4o model. Potentially more accurate/realistic than dall-e, especially with words/letters. Sora turbo: faster/smaller/less powerful version of Sora (the video model that.
4
9
73
The Manhattan Project for AI is here. OpenAI, SoftBank, and Oracle with a $500B investment will be building Stargate, a multi-gigawatt data center for AI training and inference. Adjust your expectations accordingly.
President Trump is announcing a massive AI infrastructure plan later this afternoon, and CBS is reporting it includes the return of Project Stargate. OpenAI, Softbank and Oracle are planning to invest $500 billion into Stargate over the next four years. It will be built in Texas.
3
6
75
Updated rate limits from OpenAI: . ChatGPT Plus, Team or Enterprise users have the following rate limits:. o3: 50 messages per week.o4-mini: 150 messages per day.o4-mini-high 50 messages per day. ChatGPT Pro users have unlimited access to all models.
o3 and o4-mini ChatGPT Usage Limits. ChatGPT Plus, Team or Enterprise accounts have 50 messages/week (o3), 150 messages/day (o4-mini) and 50 messages/day (o4-mini-high). ChatGPT Pro offers unlimited access to o3, o1-pro, o4-mini, o4-mini-high, GPT-4o, and GPT-4.5 models
3
10
77
With Ilya turning his attention completely to ASI it is clear that AGI is near and that it is only a matter of time.
3
8
71
Flowers confirms the search event was a bait and switch to uncover internal leakers at OpenAI.
@AILeaksAndNews They terminated at least one guy both me and Jimmy sometimes talk with, won’t tell his name obv.
2
9
74
GPT-4.5 appears to be rolling out today as a research preview, exclusively for ChatGPT pro subscribers. Interesting choice to limit it just to pro, but I am guessing they do not have the capacity to serve it to all users immediately. What do you think about this choice?.
ChatGPT web app has been updated for GPT-4.5. - "Access to research preview of GPT-4.5" feature has been added to the pricing page for Pro. - New announcement "Try the GPT-4.5 research preview - Pro users now have access to our newest, largest model.". (I was curious who would
11
0
72
Today might be the day that the American labs answer the onslaught of Chinese AI releases. Nothing has been confirmed to be released today, but OpenAI, Google, and xAI all have their next model ready to go. Which do you want most, o3-mini, Gemini 2.0 pro, or grok 3?.
14
3
73
AI releases we could potentially see today. Google.Gemini Deep Think.Gemini 2.5 Ultra.Veo 3.Imagen 4.Jules.Astra.Mariner.Smart Glasses. OpenAI.o3-pro.GPT-5 (unlikely). Anthropic.Claude 4 Sonnet.Claude 4 Opus. xAI.Grok 3.5. Meta.Llama 4 Behemoth. What are you most excited for?.
7
9
74
Llama 3.1 405B is clearly the top frontier model. Nearly everyone agrees that Sonnet 3.5 outperforms GPT-4o. So how long will OpenAI be content with being in third place?.
7
1
68
@mckaywrigley o1 pro is going to change productivity, science, medicine, math, etc . And it’s also the worst it’ll ever be.
1
1
73
If the OpenAI demo Monday isn’t for search or GPT-5, what is it for?. Will the officially release the models that have been teased on chatbot arena?. Or will they unveil something we haven’t even heard about?.
20
3
67
It’s pretty crazy that OpenAI dropped a model update just to distract us from the fact that they fired two safety researchers for raising safety concerns.
9
5
65