Matthew MacKinnon

[Matthew MacKinnon @MatthewMac69462]

@astropol0 Mini 2

[Matthew MacKinnon @MatthewMac69462]

@KLsounds @Tjernlund Almost always 1 year/room. So long enough to put some time/money into it, but certainly not this much.

[Matthew MacKinnon @MatthewMac69462]

@weswinder Perhaps it is an unfounded claim, but I would go so far as to say the free GPT-5.6 Luna Instant “hallucinates” incorrect information a little less than the average human.

So I think it can certainly be improved on, but I don’t think it’s super dumb or unusable.

[Matthew MacKinnon @MatthewMac69462]

@HSVSphere @mil000 There are 50 comments and 300 likes… I think it worked.

[Matthew MacKinnon @MatthewMac69462]

@KLsounds @Tjernlund You don’t stay in the same room all 4 years.

[Matthew MacKinnon @MatthewMac69462]

@Diivanand @cargoshortdad64 Graph is showing years of experience.

[Matthew MacKinnon @MatthewMac69462]

@ForwardEditor They should really deprecate spark.

[Matthew MacKinnon @MatthewMac69462]

What exactly did you mean by that Mosaad? 🤔

[Mosaad @mosoahmed]

@DaveShapi I love that feeling of having full token while I am in full control riding the model 😀

[Matthew MacKinnon @MatthewMac69462]

@DanDr1s This post is completely inaccurate. Did you even look at the benchmark yourself?

Only one of your statistics is correct.

[Matthew MacKinnon @MatthewMac69462]

@roanoke_gal @emollick Unbelievable that X hasn’t done this already.

[Matthew MacKinnon @MatthewMac69462]

I suspect this is gpt-image-2-mini. Not as good as gpt-image-2, but vastly superior to gpt-image-1-mini.
I doubt it will be much cheaper than the new Grok Imagine Image 2.0.
Personally, I think OpenAI should leave cheap image gen to other labs and focus on improving their SOTA.

[White @White1637402]

A new GPT image model just appeared on Arena called mona-lisa-1. It's extremely better than Image 2. 💪💪

#ChatGPT

[Matthew MacKinnon @MatthewMac69462]

@ryanbrewer Still useful when you don’t want your agent to implement anything and you’re just trying to talk to it.

[Matthew MacKinnon @MatthewMac69462]

@leviath666 @LexnLin @grok lol true. Hope I saved you some usage.

[Matthew MacKinnon @MatthewMac69462]

@timo_rf Correct me if I’m wrong, but isn’t this just codex/claude code remote control?

[Matthew MacKinnon @MatthewMac69462]

@leviath666 @grok @LexnLin @grok answer, clanker.

[Matthew MacKinnon @MatthewMac69462]

The Work & Codex documentation clearly states that Work, Codex, and Voice share one agentic usage/credit pool, and that Work uses Scheduled Tasks for recurring work. It does not explicitly say that every scheduled task, including ones created in regular Chat, are billed against that pool.
Also, it is illegal to capture the opponents king in chess. You would know that if you weren’t so retarded.

Follow the steps in my other reply to realize your fault, piggy.

@grok confirm?

[Matthew MacKinnon @MatthewMac69462]

@leviath666 @LexnLin @grok This is exactly why you are on the $20 plan, and I have 3 $200 plans. NGMI.

[Matthew MacKinnon @MatthewMac69462]

Go into the ChatGPT app on your phone. Check your usage. Open a new chat. Make sure “Chat” is selected at the top. Open the text box and type this prompt: “Create a scheduled task to run 1 minute from now to use memory and web search to diagnose my mental disabilities.”

Then send it, wait for the task to run and complete, and check your usage again. It will be the same.

[Matthew MacKinnon @MatthewMac69462]

@leviath666 @LexnLin @grok Again, you created what you thought were “scheduled tasks in ChatGPT”, but are actually codex automations, FROM THE FUCKING WORK TAB. What did you think was going to happen?