Kol Tregaskes
@koltregaskes
Followers
15K
Following
143K
Media
7K
Statuses
110K
AI, Tech & Science News Curator🔬 | AI Art Creator 🎨 | AI Video Producer 🎬🤖 | AI Music Composer 🎵 | Alt-ego of @axylusion
UK
Joined March 2008
Veo 3.1 and Veo 3.1 Fast spotted on Google Cloud Console. Ignore all the AI tools claiming they have exclusive access or making false claims about it being released in September.
13
6
159
New APIs for Sora 2, Sora 2 Pro, GPT-5-Codex, and gpt-image-1-mini and gpt-realtime-mini:
4
8
69
NeuroChat, a brain-sensing AI chatbot, introduced, boosts user engagement by adapting responses in real-time based on EEG signals. https://t.co/7n1lqF86oG Developed at MIT Media Lab as a browser-based tool using Muse 2 headband for privacy. It showed higher self-reported
Introducing NeuroChat: the world’s first neuroadaptive chatbot that adapts its responses to your cognitive engagement. By reading brain signals in real time, NeuroChat personalizes its teaching style to your attention, curiosity, and focus. NeuroChat works with your brain, not
0
0
2
No, these are not available in the UK. @OpenAI @OpenAIDevs, please clarify. I see them in the USA, though:
Apps in ChatGPT is not available in the EU yet. No news on the UK, but I cannot get them to work yet.
8
3
54
Gemini 3.0 due out on 22nd October. I don't think this was suppose to leak! Again, this date could change.
7
6
105
Epoch AI reckons GPT-5's training needed more computer power than GPT-4 but less than GPT-4.5, done by improving a base model with around 100 billion elements and more than 30 trillion bits of info after the initial run. It highlights OpenAI keeping up with competitors like
We recently wrote that GPT-5 is likely the first mainline GPT release to be trained on less compute than its predecessor. How did we reach this conclusion, and what do we actually know about how GPT-5 was trained? 🧵
1
2
15
ForecastBench update shows superforecasters outperforming GPT-4.5, with LLMs surpassing public forecasters to rank the median public at #22 on the Tournament leaderboard. The benchmark, from Forecasting Research Institute, evaluates 500 binary questions biweekly from datasets
Is AI on track to match top human forecasters at predicting the future? Today, FRI is releasing an update to ForecastBench—our benchmark that tracks how accurate LLMs are at forecasting real-world events. A trend extrapolation of our results suggests LLMs will reach
1
0
5
Apps in ChatGPT is not available in the EU yet. No news on the UK, but I cannot get them to work yet.
Apps in ChatGPT are starting to roll out today to Free, Go, Plus, and Pro users outside of the EU. More pilot partner apps and availability coming soon.
1
2
25
ChatGPT Messages (DMs) are coming? I hope we can invite friends into chats with ChatGPT.
4
0
19
Gemini 2.5 Computer Use model launched in public preview via Gemini API, built on 2.5 Pro to enable AI agents for UI interactions like clicking and typing, outperforming rivals on benchmarks like Online-Mind2Web with up to 20% higher success and lower latency. Optimised for web
💻 Introducing Gemini 2.5 Computer Use, available in preview via the API. It builds on Gemini 2.5 Pro’s vision & reasoning capabilities to power agent interactions with UIs. It completes tasks with lower latency, & outperforms alternatives on web & mobile control benchmarks.
4
10
66
Claude Opus 4.1 tops the new ASCII benchmark by @JasonBotterill3
I built a new ASCII art benchmark to compare how well LLMs create art using only text characters. So far, Opus 4.1 is the highest rated model.
0
2
20
AIVideoDotCom announces "Longshot", an AI video generator that can create a continuous 3 minute video with one prompt. https://t.co/z8Dr3WciZ0
SORA CAN'T generate a continuous 3 minute shot. No AI can. Until today...announcing “LONGSHOT”. Continuous footage up to 3 minutes long from a single prompt (prompt below).
2
4
26
Very cool.
I created a million-dollar Pixar-quality short in just 8 days using AI. Workflow: nano banana + midjourney + seedance + kling 2.5 + Suno + Elevenlabs I also tested Sora 2…and the results were surprising (see below) Excited to finally share this! 🧵
2
0
9
Qwen be busy next week.
some models next week
0
0
4
AI models improved by 20-50 percentage points on benchmarks including GPQA Diamond (+25%), Aider Polyglot (+52%), VPCT (+26%), SimpleBench (+21%), and FrontierMath (+23%) from October 2024 to July 2025. Data from Epoch’s Benchmarking Hub covers diverse skills in science, coding,
AI capabilities have been steadily improving across a wide range of skills, and show no sign of slowing down in the near term. 🧵
0
0
1
This is a big update, ChatGPT can now accept video inputs on the web. Also found in the iOS app earlier. Have anyone tried audio inputs?
8
13
148
Sorry AI community, I'm busy for the rest of the year! https://t.co/0MEnIAD1V2😜
7
1
19
