Mikito未来辉

Secret saucer pilot. Meddling in forbidden frequencies and media mortals shouldn't touch. 🛸🔮

[Flo Crivello @Altimor]

Pulled the trigger today and switched 100% of Lindy traffic to DeepSeek v4, churning from Anthropic models.
Saves us millions of $ and we're actually seeing an *increase* in performance on many core use cases. Transformative for the business.

[Mikito未来辉 @ploycatmj]

lol bytedance have so many paid plan

[BytePlus @BytePlusGlobal]

The officially launched enterprise edition of Seedance 2.0 supports multi-scene storytelling and meets complex video production needs.

[Lisan al Gaib @scaling01]

MiniMax-M3 is now the best open-source model on the AAI

It's ahead of Kimi-K2.6, GLM-5.1 and DeepSeek-V4-Pro

On the Coding Index it's slightly behind V4-Pro and K2.6, but it's beating them on the Agentic Index

CritPt scores are a bit worrisome.

But the biggest highlight here is probably its reasoning efficiency compared to other open models.
(but still miles behind closed models)

[elie @eliebakouch]

microsoft MAI tech report is a gold mine, one of the most transparent for a model at this scale.

this model uses zero synthetic data or distillation from previous models. this means reasoning, agentic behavior, tool use are all learned fully during post-training with no cold start. bold choice that makes it harder and requires more iterations to reach sota, but you get FULL control over your model series and it proves they are serious about being a frontier lab.

the tech report is insanely detailed and precise about numbers. to give an example, they give the exact MFU across all the iterations of the model, with the exact changes etc. they also share the full scaling ladder recipe, to my knowledge this is the first time i've seen this in a tech report at this scale

let's look at all of this in this likely very long thread 🧵

[Garry Tan @garrytan]

New item in my SOUL md tonight

[OpenCode @opencode]

MiniMax M3 will be launching soon

You can try it right now in OpenCode

For free

[Mikito未来辉 @ploycatmj]

they hire super dev to distill their skills my guess

[LinearUncle @LinearUncle]

这位大神开发了kitty，一个移动app可以手机上控制你电脑上的codex，CC，pi等编程agent干活，今天被openAI招聘啦！

恭喜🎉，感觉openAI很惜才，前手刚招了repoprompt的作者，现在又找了kitty的作者一起开发codex.

国内有些厂还在换语言拼cli😂，加油

[Mikito未来辉 @ploycatmj]

omo plugin is really buggy

[思维怪怪 @0xLogicrw]

在 Anthropic 随 Opus 4.8 推出 Claude Code 动态工作流与 ultracode 模式仅数小时后，开源框架 OMO 团队在 X 上发起猛烈攻势，公开指责 Anthropic 像素级抄袭了其多模型编排架构。

OMO 作为 16.7 万星开源终端 Agent 项目 OpenCode 官方认定的 No.1 插件，由 23 岁韩国黑客 Q 开发，并由其基友团队 Sisyphus Labs 宣发，目前已斩获 6 万星，在开发者社区极具声量。

这场风波把大厂「先封杀、后吸收」开源项目的惯常操作又一次暴露在阳光底下。

今年 1 月，Q 在 OMO 中推出了用于多模型动态编排的 ultrawork 工作流以及负责合并审查的协调大脑 atlas。由于这套编排机制在运行多智能体递归循环时会吞噬天量 Token，被 OpenCode 联合创始人 Dax Raad 吐槽为「Token 燃烧器」，高能耗调用直接促使 Anthropic 在 4 月份痛下杀手，强行封禁了第三方客户端使用订阅额度的通道。

仅仅三个月后，Anthropic 官方就将这套已被开源验证的动态编排逻辑收编，化身为自家 Claude Code 的闭源收费主打功能。

除了死磕 Anthropic，OMO 团队与 FactoryAI 在今年 2 月也存在旧怨。FactoryAI 此前高调宣传的 Missions 多天自主任务流，被指直接搬运了 OMO 的三层 Agent 架构：OMO 的规划器 Prometheus 变成了 Factory 的计划审批流，协调大脑 Atlas 被改名包装为 orchestrator Droid，执行器 Sisyphus Junior 变成了 Worker Droid，甚至连原装的技能过滤机制都被简化挪用。

面对 OMO 团队晒出的历史开发记录，以及要求大厂公开致谢并承认原创贡献（给点额度支持）的喊话，两家大厂至今装聋作哑。

[Mikito未来辉 @ploycatmj]

长此以往，人和人是会脱节的。墙内开的花，基于一些官方的声音。墙外是底下人的呻吟

[Mikito未来辉 @ploycatmj]

公众号上大量企业入驻，每天写出来要审稿，腾讯再过滤一遍。到你手上有时效性的东西已经经过了三双眼睛，所以真的多少赚钱技能，别人不讲的那么清楚，很多故弄玄虚需要拿出来卖。果断抛弃非个人号的阅读。x上面时效性又快，就像第一轮喂的都是优质信息。然后现在很多po主发github，又进化成白嫖圣地

[Mikito未来辉 @ploycatmj]

@lucaspatiri_ UGC

[Google Gemma @googlegemma]

Introducing the newest Coral board, for efficient, on-device AI!

Check out the demos in the video:
- On-board speech translation
- Natural language controlling hardware
- Vision & sound generating music

[Serena Ge (Datacurve) @serenaa_ge]

Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks.

On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.

[Young @Young_AGI]

Kimi Code CLI 👀
https://t.co/iAgd0DLAVi

[Mikito未来辉 @ploycatmj]

andrew dai最近的访谈中就直指deepmind和google brain之前并没有人关心语言模型和scale，都在玩自己的hobby project，最后还是gpt3出来把大家吓一跳，alex radford看出来andrew一个意外引发的街越性能提升。scale是一个工程问题，牵涉到办公室政治和credit拿结果需要上面批准，但是没结果怎么让批

[Mikito未来辉 @ploycatmj]

Saw high level language in tesla repo. U can use rust which is similar to c. For elon, he probably got fp dude

[Elon Musk @elonmusk]

SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.

The potential speed improvement vs JAX for large training runs is over an order of magnitude.

[Serenity @aleabitoreddit]

Interesting photonics selloff today on no news?

$LITE down -4.95%
$AAOI down -4.85%
$SIVE down -14.8%
$SOI down -5.73%
$AXTI down -8.13%
$IQE down -12.13%

I think it’s probably the most compelling theme going forward (even more than power semis).

Just tends to be very volatile on the way up.

Surprised about $AAOI though given there’s some institutional notes apparently about long term $AMD or $NVDA agreements. (Rosenblatt). Maybe $600m ATM caps some near term upside.

$SIVE as well, given EU Chips Act 2 is next week around photonics, and they’re listed on the blueprint. Same with MSCI/NASDAQ omx inflow next week.

I’ve been personally adding to positions since I have high conviction in the photonics theme (CPO especially) given TAM expansion overall next 2 years.

[Mikito未来辉 @ploycatmj]

grok summerise buy/hold/sell reason from gemini. $LEN

[Connor @BusDownBonnor]

This right here is what you B2B guys are missing out on…

[Mikito未来辉 @ploycatmj]

https://t.co/35OFRzhW2b covers the 150k star, "Karpathy-Inspired Claude Code Guidelines" It's made for founders. Including,
/plan-ceo-review on any feature idea
/review on any branch with changes
/qa on your staging URL