Akshat Bindal

Exploring new Gen AI models and tools. Current fav: Clause opus 4.8 & NotebookLM

[Akshat Bindal @bindal_akshat]

@AjaySinghadiya8 yes, I thought it was gradual rollout initially and then they rolled out for all users, but seems like it was not the case.

[Akshat Bindal @bindal_akshat]

Future of voice dictation apps:
https://t.co/45rRGFnyYt

[Akshat Bindal @bindal_akshat]

@xikhar great work. So cool. Can we resize it to be a bit smaller?

[Akshat Bindal @bindal_akshat]

@AnthropicAI I agree. Just don't go with the flow that everyone in industry is signing the letter and council, so they are right. Dario's points are so valid that open source doesn't mean it will help defenders more than attackers, the opposite seems more probable to me.

[Akshat Bindal @bindal_akshat]

Why can't gemini app on web and on mobile edit my docs/slides/sheets directly?
Right now it generates the reports in canvas and when I ask it to update respective doc, it just export it into a new doc. You guys have this kind of ecosystem, still lacking basics? @joshwoodward

[Akshat Bindal @bindal_akshat]

@LyalinDotCom Why can't gemini in chrome take actions then? You own chrome, you own gemini. You don't need any extension. Go ahead.

[Akshat Bindal @bindal_akshat]

@joshwoodward I have been doing that from last year. Quite cool.

[Akshat Bindal @bindal_akshat]

Yesterday the knowledge cutoff date for the @GeminiApp models were showing as mar26, but today they have made it unknown😂
Why so @OfficialLoganK ?

[Akshat Bindal @bindal_akshat]

Hi @antigravity team,
Your model picker looks really ugly in Antigravity. Can you please vibe code a new model picker for it?
Just simply moving the thinking effort apart from model would also work.
Also the 3rd party models are outdated, kindly refresh them or remove them.

[Akshat Bindal @bindal_akshat]

@OfficialLoganK You guys are so cooked man, now you are not even comparing your new model with competitors on the scorecards😭..Comeback when?

[Akshat Bindal @bindal_akshat]

@joshwoodward @OfficialLoganK Please launch something, its getting boring here at https://t.co/tySEytx1qi

[Akshat Bindal @bindal_akshat]

@GoogleAIStudio https://t.co/spsu84YZH4

[Akshat Bindal @bindal_akshat]

@future_realms @GoogleAIStudio Trolling level >>>> 😂

[Akshat Bindal @bindal_akshat]

A few scope of improvements:
-Can add a dynamic thinking mode (Auto mode) by which instead of user specifying the thinking effort, it automatically gets based on the task, how much effort to use.
@ChatGPTapp

[Akshat Bindal @bindal_akshat]

GPT 5.6 is absolutely beast!
Here's what I liked about it and what I built with it:
- luna, terra, sol : Pure bliss naming convention.
- The in-app browser, made web dev so enjoyable.
- The new Model picker is too good.
- The unified app : chatgpt work + codex = 🫶
@ChatGPTapp

[Akshat Bindal @bindal_akshat]

With GPT 5.6, Built a real-time AI ops center for FIFA 2026 ⚽🏟️
→ GenAI incident triage & dispatch → 82K fan crowd management → Weather-adaptive egress routing → Multilingual AI concierge → Live audit trail + volunteer roster
Absolute Beast model!
🔗https://t.co/bd9sxhgnyT

[Akshat Bindal @bindal_akshat]

@demishassabis Well thought. Just a suggestion regarding the last point of expanding it to international leve later on, can't we start building it with representatives from every country from the very beginning only?

[Akshat Bindal @bindal_akshat]

And gemini is even worse with naming, with preview & GA version on top of all the mess. They can do like:
Gemini 4 (low, med, high)
Nano Banana 3 (low, med, high)
Omni 2 (low, med, high)
Lyria (low, med, high)
Gemini Live 3(low, med, high)
Gemma 4 (low, med, high)
Easy-peasy...

[Akshat Bindal @bindal_akshat]

Question: Why don't companies just unify their different models into a single model with just 3 level thinking efforts, for example, instead of claude haiku 4.5, sonnet 5, opus 4.8, fable 5, they just do it: Claude 5 (low, med, high), it's so damn easy!
Updates could be 5.1, 5.2

[Akshat Bindal @bindal_akshat]

OpenAi tried the same with GPT 5.6 by naming it luna, terra, sol, but they still have low, med, high, xhigh, ultra, pro inside each of these, WTH, if the models are so intelligent, then they should know how much to think for a task, not depend on anybody specifying it manually..