Specter👁 (@0xSpecter_) header
Specter👁 (@0xSpecter_) profile picture

AI researcher

My telegram channel ⇢
Joined July 2022
Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)Photo by Specter👁 (@0xSpecter_)
Specter👁 retweeted
It seems that we are gradually approaching the moment when a huge studio and a team of hundreds of developers will no longer be needed to create our own GTA Matt Schumer showed what can be done with AI agents and Claude Fable 5.2. In just 16 hours, he assembled an early prototype of an open world game - photorealistic New York, through which you can move right in your browser But it's not even the result itself that's more interesting, but the speed with which it was obtained Over the past month, the developer has spent more than 400 billion tokens experimenting with the model and learning how to manage AI agents so that they can independently create complex game systems Of course, it's still a long way to a full-fledged GTA. But the very fact that one developer can assemble something similar in 16 hours makes you think Perhaps, soon the main limitation in the creation of games will no longer be the size of the team and the budget, but only the scale of your idea and the ability to properly manage AI
2
1
10
310
It seems that we are gradually approaching the moment when a huge studio and a team of hundreds of developers will no longer be needed to create our own GTA Matt Schumer showed what can be done with AI agents and Claude Fable 5.2. In just 16 hours, he assembled an early prototype of an open world game - photorealistic New York, through which you can move right in your browser But it's not even the result itself that's more interesting, but the speed with which it was obtained Over the past month, the developer has spent more than 400 billion tokens experimenting with the model and learning how to manage AI agents so that they can independently create complex game systems Of course, it's still a long way to a full-fledged GTA. But the very fact that one developer can assemble something similar in 16 hours makes you think Perhaps, soon the main limitation in the creation of games will no longer be the size of the team and the budget, but only the scale of your idea and the ability to properly manage AI
2
1
10
310
Specter👁 retweeted
Google has a new multimodal model - Gemini Decisions She understands text, video, audio and images. And all this in one model! A million input tokens costs only $0.10, and you can try it for free at AI Studio
1
2
16
435
Google has a new multimodal model - Gemini Decisions She understands text, video, audio and images. And all this in one model! A million input tokens costs only $0.10, and you can try it for free at AI Studio
1
2
16
435
Specter👁 retweeted
OpenAI can release GPT-6.2 Astra on November 19 According to rumours, the company is deliberately holding the release to present the model right before the Anthropic IPO It seems that OpenAI wants to draw the market's attention to itself at the most inappropriate moment for the competitor If it's true, the timing will be quite interesting
2
1
9
470
OpenAI can release GPT-6.2 Astra on November 19 According to rumours, the company is deliberately holding the release to present the model right before the Anthropic IPO It seems that OpenAI wants to draw the market's attention to itself at the most inappropriate moment for the competitor If it's true, the timing will be quite interesting
2
1
9
470
Specter👁 retweeted
Google rolled out Gemini 4 Carbon This is a model for tasks where AI needs to think for a long time, dig into the code, analyse files and check different hypotheses On the MKA test, she scored 54%, slightly losing to Fable 5.2 Max with 57% Already available in the API and for Ultra subscribers. Until the end of November, prices are halved, then it will be more expensive
1
1
14
522
Google rolled out Gemini 4 Carbon This is a model for tasks where AI needs to think for a long time, dig into the code, analyse files and check different hypotheses On the MKA test, she scored 54%, slightly losing to Fable 5.2 Max with 57% Already available in the API and for Ultra subscribers. Until the end of November, prices are halved, then it will be more expensive
1
1
14
522
Specter👁 retweeted
Someone's bread seems to have been taken away
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data. Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
1
1
342
Someone's bread seems to have been taken away
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data. Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
1
1
342
Specter👁 retweeted
OpenAI added businesses to ChatGPT Now companies can create their own AI agents that will answer customer questions, show products and help with the choice In fact, ChatGPT is becoming another sales channel for business
1
1
6
364
OpenAI added businesses to ChatGPT Now companies can create their own AI agents that will answer customer questions, show products and help with the choice In fact, ChatGPT is becoming another sales channel for business
1
1
6
364
Specter👁 retweeted
Now Codex can create games for Game Boy OpenAI added the ModRetro Chromatic Codex Edition retro console to its store for $200 You can describe the idea of the game to Codex, and it will collect levels, add animations and prepare ROM. The finished game can be run in a browser or burned to a cartridge and play on a real console For those who want to try, OpenAI has also posted a gallery of games, including FlapGPT
1
2
10
555
Now Codex can create games for Game Boy OpenAI added the ModRetro Chromatic Codex Edition retro console to its store for $200 You can describe the idea of the game to Codex, and it will collect levels, add animations and prepare ROM. The finished game can be run in a browser or burned to a cartridge and play on a real console For those who want to try, OpenAI has also posted a gallery of games, including FlapGPT
1
2
10
555
Specter👁 retweeted
F*ck, how to comment on it?
Created with Fable 5.2 by an Anthropic engineer. THIS IS AAA-LEVEL STUFF!!!!!!!
2
5
288
F*ck, how to comment on it?
Created with Fable 5.2 by an Anthropic engineer. THIS IS AAA-LEVEL STUFF!!!!!!!
2
5
288
Specter👁 retweeted
Anthropic is preparing for the IPO on Nasdaq. According to rumours, the company may start meeting with potential investors on November 10. If the plans are confirmed, the IPO may be among the largest in the history of the stock market.
Anthropic presented Fable 5.2. The new model works 30% faster than the previous version, and the cost of tokens has decreased by 20% and is now $8/$40 per million input and output tokens. Fable 5.2 is already available via API and in the Claude subscription. Even Pro users will have access to it in the first week. For Max, Ultra and Team subscribers, they also provided from 1 to 3 additional limit resets depending on the subscription period. Pro users got one reset.
1
2
14
467
Anthropic is preparing for the IPO on Nasdaq. According to rumours, the company may start meeting with potential investors on November 10. If the plans are confirmed, the IPO may be among the largest in the history of the stock market.
Anthropic presented Fable 5.2. The new model works 30% faster than the previous version, and the cost of tokens has decreased by 20% and is now $8/$40 per million input and output tokens. Fable 5.2 is already available via API and in the Claude subscription. Even Pro users will have access to it in the first week. For Max, Ultra and Team subscribers, they also provided from 1 to 3 additional limit resets depending on the subscription period. Pro users got one reset.
1
2
14
467
Specter👁 retweeted
Anthropic presented Fable 5.2. The new model works 30% faster than the previous version, and the cost of tokens has decreased by 20% and is now $8/$40 per million input and output tokens. Fable 5.2 is already available via API and in the Claude subscription. Even Pro users will have access to it in the first week. For Max, Ultra and Team subscribers, they also provided from 1 to 3 additional limit resets depending on the subscription period. Pro users got one reset.
2
2
1
18
918
Anthropic presented Fable 5.2. The new model works 30% faster than the previous version, and the cost of tokens has decreased by 20% and is now $8/$40 per million input and output tokens. Fable 5.2 is already available via API and in the Claude subscription. Even Pro users will have access to it in the first week. For Max, Ultra and Team subscribers, they also provided from 1 to 3 additional limit resets depending on the subscription period. Pro users got one reset.
2
2
1
18
918