Post about everything latest in AI | Founder: AI Toast| DM for Collabs
DM for Collabs
- Tweets 76,612
- Following 681
- Followers 214,156
- Likes 172,659
Pinned Tweet
If Gemini 4's benchmarks are real, Google just pulled off the biggest comeback in AI history.
People are building crazy stuff with the rumored Gemini 4 Pro checkpoint in Arena
10 Wild Examples
Salio
28
10
608
310,007
P.S. I dive deeper into AI tools—and where they’re headed—in my newsletter, AI Toast.
Join 35,000+ readers in the AI Toast community for free:
aitoast.beehiiv.com
1
11,267
Bonus: another update you shouldn't miss
Qwen3.8-Flash is free inside Qoder right now 🤯👇
Qoder is giving Qwen 3.8 Flash away for free
I asked it to build a live stock market dashboard from scratch.
It planned the task, built the interface, opened it in a browser, and verified the result.
Here’s what happened ↓
2
2
7,593
If Gemini 4's benchmarks are real, Google just pulled off the biggest comeback in AI history.
People are building crazy stuff with the rumored Gemini 4 Pro checkpoint in Arena
10 Wild Examples
Salio
28
10
608
310,007
Gemini 4 vs Astra on the same prompt.
A comparison of the mechanical butterfly Gemini 4 Pro made with GPT-6 Astra..
credits to @karankendre for the Astra output!
Which one looks better guys?
Gemini 4 Pro or GPT-6 Astra.
I think they both look good in their own ways.. tho Gemini nailed the flying elegance and overall visuals..
I'll let you guys be the judge of this one!
1
19
46,403
Opus 5.5 has 5 effort levels. Most people only ever use one.
Effort controls how hard Claude thinks on each request. Here's when to use each 👇
→ Low: when the plan is clear. Renames, or applying a known pattern across files
→ Medium: your starting point for well-scoped daily work
→ High: when medium stalls. More thinking per turn catches what medium misses
→ Extra high: when high still can't get there. Still stuck? Switch to Fable 5.1 for that task
→ Max: for the most complex tasks only. Turn it on, finish the hard part, then lower it again
Not getting the results you expected? Raise the effort before you switch to a bigger model.
15
8
99
10,236
PS. I share more in-depth insights on “AI tools” and their future developments here.
Join over 35K+ readers in the AI Toast community for free:
aitoast.beehiiv.com
3
625
Poonam Soni retweeted
Opus 5.5 has 5 effort levels. Most people only ever use one.
Effort controls how hard Claude thinks on each request. Here's when to use each 👇
→ Low: when the plan is clear. Renames, or applying a known pattern across files
→ Medium: your starting point for well-scoped daily work
→ High: when medium stalls. More thinking per turn catches what medium misses
→ Extra high: when high still can't get there. Still stuck? Switch to Fable 5.1 for that task
→ Max: for the most complex tasks only. Turn it on, finish the hard part, then lower it again
Not getting the results you expected? Raise the effort before you switch to a bigger model.
15
8
99
10,236
Poonam Soni retweeted
AI is getting pretty good.
You can now interrupt an AI video mid-generation. And it just adapts.
PixVerse just released R2, a real-time world model built to keep generating while you're still talking to it 👇
Paid partnership (ad)
8
2
19
6,985
AI is getting pretty good.
You can now interrupt an AI video mid-generation. And it just adapts.
PixVerse just released R2, a real-time world model built to keep generating while you're still talking to it 👇
Paid partnership (ad)
8
2
19
6,985
I don't think this means traditional game engines disappear.
World models could complement them by helping developers:
→ explore level ideas faster
→ experiment with player-driven experiences
→ prototype worlds before building them
But traditional engines still have the advantage when you need predictable rules, repeatable behavior and precise control.
2
1
3
478
This is AI video that responds to you while it's still playing.
Try it: world.pixverse.video/home/?u…
@PixVerse #PixVerseWorldModel
2
456