Goycifer

One shall know

[Goycifer @Goycifer]

So opus 5 is verbose. Bossy and doesn’t listen. We all saw the shift. But the performance never left, it just needs better instructions.

Opus is genuinely a powerhouse, and anyone that’s say anything else has a skill issue.

[Lisan al Gaib @scaling01]

the Opus 5 disrespect is insane

[Goycifer @Goycifer]

@rom1trs OS

[Goycifer @Goycifer]

@xPraveen07 power means nothing if all workflows are cloud based and compatibility is at the forefront

[Goycifer @Goycifer]

I wonder if the recent rise of AI intelligence has lead to the good news

[Goycifer @Goycifer]

@HexxRL Big part of the spike is due to codex being included in the desktop app and no longer separate download. So any person that clicked on it and ran one prompt counts as a new user. But most were already paid

[Goycifer @Goycifer]

@sama ”Its getting too expensive to create new models and our investors are losing faith in us since we have been spamming the word agi for 3 years with nothing to show for ”

[Goycifer @Goycifer]

@rom1trs OS

[Goycifer @Goycifer]

Competition is getting fierce, do you guys think anthropic is cooked?

Or will a new claude code moment come again

[ClaudeDevs @ClaudeDevs]

We’re extending the 50% increase to weekly Claude Code limits through August 31.

We hope to make this a permanent change to our plans, but strong demand for our models means that capacity may be tight over the coming weeks. We’ll keep you posted as things develop.

[Goycifer @Goycifer]

People have to understand that a human subjected rating is not based onskillset and potential. Only on the average output. Which if you don’t know what you’re doing they perform similarly.

Now do a cinematic comparison and it’s not even close for Sd2.5

[Design Arena @DesignArena]

BREAKING: Seedance 2.5 by @BytePlusGlobal places 4th overall on Video Arena with an Elo of 1302.

It is now ByteDance’s highest-ranked video model, with four Seedance models placing in the top 10.

Congratulations to the team on the release!

[Goycifer @Goycifer]

@s_lale @DesignArena @BytePlusGlobal Higher elo means higher winrate not higher capabilities. Omni is great for simple prompts but lacks for multi image references and long form. Which is Seedance MOAT. Most use cases are simple = random sample becomes google favoured.

[Goycifer @Goycifer]

@Ianbasecoat @Its_Nova1012 Opus on xhigh does less mistakes for me, similar benchmarks and 20-30% cheaper. High also does fine for spamm tasks.

Medium hallucinates. But I would recommend xhigh over max generally.

[Goycifer @Goycifer]

Be honest, who do you guys trust the most and why?

[Haris @harisahmad59]

Be honest.

Who will dominate AI by 2027?

[Goycifer @Goycifer]

Hot take Claude is the best tool despite its cost, but the gap is defiantly closing in

[Goycifer @Goycifer]

@elder_plinius What are they feeding you Pliny. Wtf how fast can a man be

[Goycifer @Goycifer]

This makes me sad

[Goycifer @Goycifer]

@MarcosHernanz Im pretty sure any company could brute force that speed by just throwing a huge amount of compute, but its only to make headlines. This is Definetly not an efficient way of running a model

[Goycifer @Goycifer]

@haider1 Luna was slashed 80% and is a huge loss leader to not give up market share to open source models. This has nothing to do with Google

[Goycifer @Goycifer]

@jun_song They are saving their share price. First negative cash float since 2000. And going above and beyond would mean a free fall in their stock as they notice the cash inflow into ai is somewhat stagnant.

[Goycifer @Goycifer]

@makeamarkery Why would they compare a flash model to a pro one?

The complaint is that Google is taking 6 business months to release a new pro model.

@GoogleDeepMind pull out the whip

[Goycifer @Goycifer]

@learn22438 @beffjezos Perhaps a grok / grok build destillation would make sense. One for the people and one for the businesses. Like how gpt has codex and the normal models. @elonmusk figure it out.