Artificial Intelligence thread

[meedicx]

https://twitter.com/i/web/status/2051313629769675048

Doubao/ByteDance is starting payment plan options on its very popular app.

Doubao haven't started subscriptions yet. This is just a patch note in an iOS update. I predict they will roll it out with an updated Seed model and introduce a coding plan as well. Bytedance has been investing in building TRAE - their own AI coding IDE and harness.

There is a rumor going around that an upcoming Chinese model benchmarked really well on a widely followed private benchmark. Since Seed 2.0 was a while back, Bytedance is the only major lab without a release in the last 2 months. Also Bytedance charging such a high price surprised many people, but would actually be very cheap if it is a SOTA model with a coding plan.

 

[jnd85]

The whole Vibe Code thing just the hysteria in the bubble. Happens in every new technology. The problem if was valued at the right price there will be no bubble but these CEOs had hyped this tech to the moon and that is the problem, now they have to deliver. But at the cost of electricity increase and the cost serving these models become higher with demand I think these models are not going deliver what they are going to be priced, specially when there are cheaper Chinese models alternatives and even those can see their costs increased.

I VibeCoded myself for my own applications, I used Claude, I used OpenAi, now I mostly use my own AI Server and local models, because they meet my demands. Do the slop they produced worth it? Yes, for some applications, It worth the trillion dollars that these CEOs are hyping for? Absolutely no.

View attachment 174424

Definitely agree with all above about AI being a bubble, and about the benefits being real, but not with respect to the cost. The return on investment only makes sense when the costs are completely hidden.

Sadly, I think a lot of "progress" is just different scams wrapped in shiny packaging, like street performers who distract you while their friends pick your pockets.

 

[siegecrossbow]

Please, Log in or Register to view URLs content!

 

[bsdnf]

Doubao haven't started subscriptions yet. This is just a patch note in an iOS update. I predict they will roll it out with an updated Seed model and introduce a coding plan as well. Bytedance has been investing in building TRAE - their own AI coding IDE and harness.

There is a rumor going around that an upcoming Chinese model benchmarked really well on a widely followed private benchmark. Since Seed 2.0 was a while back, Bytedance is the only major lab without a release in the last 2 months. Also Bytedance charging such a high price surprised many people, but would actually be very cheap if it is a SOTA model with a coding plan.

View attachment 174421

Seed2.0's performance was mediocre, but after the impact of Seedance2.0, no one will underestimate ByteDance anymore.

 

[Wrought]

Judicial paperwork is a low-hanging fruit for productivity gains.

In a social media post on Tuesday, the Shenzhen Intermediate People’s Court said each judge in the city had handled an average of 744 cases in 2025 – up by 249 cases from the previous year. That made Shenzhen’s judges the most efficient and productive in Guangdong, since they handled 261 cases more than the provincial average, it said.

The average number of cases handled by judges in Shenzhen last year was also about twice the national average of 354 cases reported in 2024 by People’s Court Daily, the official publication of China’s supreme court. Zhang Jun, president of the Supreme People’s Court, told the national legislative meeting in March that Shenzhen’s AI system trial had produced “excellent results”, according to the social media post. He said the pilot scheme had now been rolled out across 23 intermediate and local courts in 11 provinces using an AI system based on the one used in Shenzhen.

Please, Log in or Register to view URLs content!

 

[iewgnem]

The whole Vibe Code thing just the hysteria in the bubble. Happens in every new technology. The problem if was valued at the right price there will be no bubble but these CEOs had hyped this tech to the moon and that is the problem, now they have to deliver. But at the cost of electricity increase and the cost serving these models become higher with demand I think these models are not going deliver what they are going to be priced, specially when there are cheaper Chinese models alternatives and even those can see their costs increased.

I VibeCoded myself for my own applications, I used Claude, I used OpenAi, now I mostly use my own AI Server and local models, because they meet my demands. Do the slop they produced worth it? Yes, for some applications, It worth the trillion dollars that these CEOs are hyping for? Absolutely no.

View attachment 174424

Vibecoding can be extremely powerful if you have the right setup

The problem with Claude and OpenAI is, and this is after I used both Claude and Kimi, that there is zero way they can justify their valuation when their capabiltiy and quality is about as differentiable from free open source models as wine tasting.

 

[9dashline]

Vibecoding can be extremely powerful if you have the right setup

The problem with Claude and OpenAI is, and this is after I used both Claude and Kimi, that there is zero way they can justify their valuation when their capabiltiy and quality is about as differentiable from free open source models as wine tasting.

opencode + kimi ?

 

[meedicx]

Seed2.0's performance was mediocre, but after the impact of Seedance2.0, no one will underestimate ByteDance anymore.

Seed 2.0 is very mediocre in coding (which they admit) but is SOTA on image recognition, which they spend all their focus. Get it to count things in images; it performs way better than any other model including Gemini.

Like Gemini, they probably worked on it before agentic coding became the hot thing. I expect the next release for both Gemini and Seed to have a huge jump in coding capabilities.

 

[temporary1]

Vibecoding can be extremely powerful if you have the right setup

The problem with Claude and OpenAI is, and this is after I used both Claude and Kimi, that there is zero way they can justify their valuation when their capabiltiy and quality is about as differentiable from free open source models as wine tasting.

Claude is obvious better in real world use case for coding.

Not sure with Opus 4.7 and their adaptive thinking bs though, but if you use Codex with gpt-5.5 xhigh, nothing comes even close to that. Absolutely smashes through everything

But regardless, good setup is always needed of course (note that gtp-5.5 is imo less reliant on good setup than 5.4)

 

[jli88]

The whole Vibe Code thing just the hysteria in the bubble. Happens in every new technology. The problem if was valued at the right price there will be no bubble but these CEOs had hyped this tech to the moon and that is the problem, now they have to deliver. But at the cost of electricity increase and the cost serving these models become higher with demand I think these models are not going deliver what they are going to be priced, specially when there are cheaper Chinese models alternatives and even those can see their costs increased.

I VibeCoded myself for my own applications, I used Claude, I used OpenAi, now I mostly use my own AI Server and local models, because they meet my demands. Do the slop they produced worth it? Yes, for some applications, It worth the trillion dollars that these CEOs are hyping for? Absolutely no.

View attachment 174424

Anthropic is growing like > 10x every year in annualized revenue. Most developers that I know are using it extensively and are impressed with it, that the growth can continue for quite a while.

This is like SpaceX all over again, Chinese analysts were in denial until very recently about the feasibility of the tech.

China needs to improve its innovation ecosystem where high risk bets with large capital can be placed.