Deep Insight Labs retweeted
Here's a serious and thorough post about the state of affairs at @Kimi_Moonshot and more importantly the possible causes. This is a long post, built on more than a decade of experience.
I have been roasting them for the past two days, but a month ago I also defended them when Anthropic said they routed Kimi user requests to Opus - see my Articles.
First, let's start with an overview of what Moonshot has done to its users over the past 6 months:
- Cut Kimi plan quotas so bad after the K2.7 and K3 releases that they're all unusable and basically scams. OpenAI's cuts are bad; Kimi's are twice as bad
- Released K3 in July with a campaign of unmarked, sponsored influencer post with faked videos of what their model had supposedly produced - I called it out then, too
- Price fixed Kimi K3: inference providers in the US and EU are not allowed to set their own prices but must follow the minimum prices set by Moonshot. This is illegal in every functioning country; Apple got convicted for this with their ebook pricing, remember?
- Stole the open source code from Cua, relabelled it and claimed they had invented computer use in their Kimi agent
- Released K3.1 with another unmarked influencer campaign, with more fake videos
- Tampered with K3.1 benchmark scores for DeepSeek in their release materials
When you look at the above, you see that these are pretty serious things that could break any brand. If you do this, you either have to think your users are stupid, or not care, or be stupid yourself.
When it comes international brands selling into China, and Chinese brands selling to international markets, they both exhibit all three characteristics.
I've heard to many Europeans talk about their business ideas: just pick a product, slap a European-brand on it, talk about it being from whichever country, and then the Chinese customers will just buy it without thought.
This never works, but these entrepreneurs are ignorant, have no will to learn the market, and regard their potential customer base as NPCs. They view China as another world and cannot reason about it, and because they cannot reason about China, they cannot treat it seriously.
For this reason, there have been numerous advertising scandals with European brands getting slaughtered online, because their leadership in HQ outside of China made tone dead decisions in their communication. D&G, Prada, Nike, adidas, Arcteryx, there's been too many internet campaigns against brands for their ad campaigns to remember.
For most Chinese companies trying to go international, it is the same but in reverse. They do not understand the international markets, they don't care to educate themselves, they see the world outside China as something otherly, and feel shielded from any negative repercussions from far away countries. They see their users as NPCs.
If you wanted to get a refund for your Kimi coding plan, what can you really do? Most likely nothing, and your X posts won't get noticed.
When it comes to the campaign for the K3.1 launch, Moonshot did a number of things that just spell incompetence:
- Create ready made videos for publishing by influencers, this is basically unheard of
- Try to hide that posts are sponsored, to generate the impression of organic user feedback
- Require approval before publication
When you do these kinds of campaigns, you don't ask for approval because you need to trust the creator; otherwise the content suffers.
You don't try to hide that videos are sponsored, because the potential backlash for this type of manipulation is 100x worse than the value, as we see now.
And well, creating fake videos.... 无语。
Some have speculated that the error is on the side of a marketing agency, but that is unlikely. The brief is always created by the company for the agency, and everything that goes out is approved by the customer.
Most likely, the marketing team at Moonshot must have said that doing hidden influence campaigns like this is risky, but could have been forced by leadership who wanted to "ensure ROI" on their marketing investment by controlling content.
So what it really comes down to is a company leadership that does not know or care about laws in other markets or does not care about them, feel like this type of manipulation is worth the risk, and see users as NPCs.
At the end of the day, this is false advertising and when you combine this with Moonshot's price fixing, it is not just that it is dishonest but it is complete and utter disrespect of laws of other countries - mind you, all of this is certainly illegal in China, too.
To close this out, what really needs to happen now is that the price fixing is reported and investigated. Inference providers are clearly not able to push back on Moonshot themselves, as they're in a competition to be able to offer the newest hyped model as soon as possible.
Don't ban the models, but hit down hard on illegal business practices by the company.
Deep Insight Labs retweeted
That's how Kimi AI are promoting their new model Kimi K3.1 pro and I know all the influencers working with them and none of them are honest,I have posted about it myself but I didn't join this and never used their templates that they provide.
PROPAGANDA, I have more to share it's f**cked up I'm gonna have follow up for this
Deep Insight Labs retweeted
if you look at the replies and reposts of the main post AND the paid partnership posts most of them are fake and just bot farms at large scale
so against X ToS and quite easy to cluster since it's always the same accounts (there is a good blog post to make with "we asked grok 4.7 to identify all the clusters and ban the bot farms")
jev used this kind of slop agency and it definitely works and converted into real usage (and they raise a massive amount with this) so i can understand why people are doing it, it's just sad and should be fixed at the platform level quick otherwise everyone will do it and people will start to leave X
i find it really sad that X is becoming a pay to win platform and that even "good" releases are going through these fully slop/fake campaigns to get traction
i'm also pretty sure there is very shady stuff where they are boosting each other's posts with bot farms and more. would be great to investigate here @elonmusk and make sure the algorithm doesn't push their posts, you can very easily cluster that if you look at each release and the accounts amplifying 🙏
(not a dunk on recursive, recently a lot of people are doing that unfo and it's definitely working, posts often get over ~1M views)
Deep Insight Labs retweeted
A fundamental belief of mine is that we are entering the age where quality of work will matter more than anything else
Historically, producing any sort of work, whether a service or product, was always about finding a balance between quality and speed
Almost always, you had to trade one for the other in order to hit certain timelines, milestones, or to keep promises you made to your customers
With AI, that constraint is slowly going away. Speed is rarely the bottleneck for high quality anymore. Work that used to take days can now be done in the span of hours across many white collar professions
The result is that speed is no longer impressive because everyone has access to it
However, what these AI tools don’t give you is quality of work product. Much like anything else, these tools only produce quality at the level of the person using the tool.
Just as an example, someone who has never coded in their life cannot produce the same quality of software that an experienced engineer using AI can create
To this day, work completed using AI remains very distinguishable vs work created with human expertise behind the wheel guiding the AI
As AI only created products and content exponentially increase, people will begin to search for human quality in the midst of endless LLM generated alternatives
This will the deciding criteria for many choices that the every day consumer will
make in the future. Build and create accordingly
Deep Insight Labs retweeted
Introducing the Mike Workflows repository:
MikeOSS is a fully open source legal AI platform, where the code is freely accessible to anyone. That is the key differentiator from proprietary platforms that are valued in the billions.
What those platforms market heavily as well are the proprietary skills and workflows that their legal engineers build for their platforms and they sell to their clients.
A while ago, I announced that I will be building a repository of workflows for legal work that will be open source and platform agnostic.
Today, I have an MVP version of that ready, the product of contributions from professionals from around the world and a variety of backgrounds, ranging from biglaw, tech and government. The workflows follow the Skill.md standard and thus you can use them in MikeOSS, ChatGPT, Claude or any other platform of your choosing.
Thank you all who have contributed and contributions are always welcome.
github.com/open-legal-produc…
Steven Sinofsky on why it's so hard to diffuse AI into companies: most work runs on know-how that's hard to turn into step-by-step instructions for an agent.
"Algorithmic thinking is really, really, really hard for the vast majority of people who have jobs."
"If you were to go to any person and ask them to create a flow chart for a particular thing that they have to go do, they would probably fail at producing that flow chart."
"Say doing a marketing plan and there's 50 marketing people working on a giant product line. One person probably understands and could document the flow chart."
"So if you put one of these agents in front of people to create these things, their ability to explain to it what to do is really, really limited."
"You're basically just developing the next abstraction layer for how people interact. And developing an abstraction layer has historically, at each level of the abstraction layer, been a highly skilled, very specific individual within an organization."
@stevesi
Jean-Denis Greze, founder of @TownAI, argues AI isn't diffusing because most people aren't tinkerers: a text box isn't enough, so the product must both identify the value prop and perform the work for them.
"There's this business that rents surfboards on an island to people and then uses Town to inspect surfboards and little beach buggies, whether they've been damaged, to decide whether afterwards the person should get their deposit back."
"What makes it work there is the tinkerer... The unlock is, is there someone over there that gets AI enough to wanna tinker a solution to their problem?"
"And the problem with that is that there aren't enough tinkerers in the world, or there aren't enough people that have the time to be tinkerers to help their business. And I think that's why AI is not diffusing."
"Instead what you're doing is you're using [AI tools] as a replacement, as like a better Google search."
"If you can make a non-tinkerer into getting the same value as a tinkerer... you could be a ginormous business today because AI can solve so many business problems. You don't need more intelligence. The intelligence is there."
@jgreze @arampell
Deep Insight Labs retweeted
An investment banker, private equity investor, venture capitalist and operator all walk into a NYC bar
The owner brings out drinks for each of them
The investment banker takes a sip and tells the bartender that he can get the owner 14x EBITDA for the place. The owner says he is not interested. The banker then slides him a business card and says to call him when he is ready
The private equity investor tells the owner he already has financing lined up to buy the bar. He can cut in the owner for a 20% earn out, but will probably fire half the staff. He points to the vodka bottle prices and notes a lot of room here for “margin expansion”
The venture capitalist, meanwhile, asks the owner if he has thought of incorporating AI into the drink pouring process. He advises the owner to think from first principles and claims he can write him a check at a $2B valuation. The only condition? The owner must convince A16Z or Sequoia to invest first.
The operator is the only person who figures out how the bar functions, where kegs are stored and how the employees are paid
At the end of the night, the bill arrives and the operator is forced to pick up the tab since he is the only one who remained sober. He asks his friends to pay him back
The banker offers to handle the payments in exchange for a fee
The private equity guy asks him if he can leave the money as owed debt, and he will pay with interest over time
The venture capitalist tells him to mark up the value of the tab but says he can’t pay it for at least another five years.
The next morning, the operator wakes up to 3 LinkedIn posts from each of his friends on what it takes to run a successful bar in NYC
Deep Insight Labs retweeted
Not surprised, data will become a tradable commodity and RL builders will be specialized refineries, the application layers are the gas stations.
AI bubble world just moved 100x faster than the real world, it seems like you’re moving slowly when you’re scrolling X.
You’re not, you’re so far ahead just knowing about this industry
Deep Insight Labs retweeted
E2B's founder @mlejva just dropped a banger essay: "Sandboxes are dead."
Coming from the CEO of the literal AI Sandbox king, 2B+ sandboxes spun up, 10M+ monthly SDK downloads, this sounds insane.
But if you look at the brutal unit economics and tech warfare behind E2B, you realize it's not a hot take. It's a survival play. 🧵👇
In his post, Vasek says competing on "cold-start milliseconds and price per vCPU-hour" is the wrong answer for 2026. Instead, he's pitching an "agent-driven private cloud" that scales on the fly.
Why is the market leader trying to kill its own category? Because pure compute sandboxes are getting squeezed from both ends.
1/ E2B's biggest moat is being hijacked.
By open-sourcing their Firecracker infra under Apache-2.0, E2B became the industry's default socket.
The dark side? Big customers like Manus just self-host E2B to avoid big bills.
Worse, giants like Perplexity/Tencent launched faster startup clones that brag about "zero-code migration, just swap the E2B URL env var."
When rivals use your own SDK to steal your cloud margin, basic VMs become a commodity trap.
2/ The billing Achilles' heel
Agents spend ~90% of their lives waiting on slow LLM tokens or human replies. E2B bills for full duration, you pay for every second the VM is alive, even at 0% CPU.
Rivals like Vercel Sandbox now use "Active-CPU metering," stopping the clock during I/O waits. If an agent idles, E2B's bill can bloat 6x–10x higher than competitors.
Moving away from "price per vCPU-hour" changes a conversation E2B doesn't want to have.
3/ The Magical Contrast
Look closely at Vasek's checklist for what production agents need: "CPU and memory that scale on the fly."
The irony? Right now, E2B can't even do live vertical resizing. CPU and RAM are hardcoded at template build time. If an agent hits an 8GB data spike, it OOMs and dies. Plus, E2B is still CPU-only, zero GPUs.
With rivals like OpenComputer (live resize) and Modal (serverless GPUs) circling, Vasek is turning his biggest engineering gap into his next grand narrative. Classic Silicon Valley judo.
4/ What can't be cloned? Data.
Anyone can fork Firecracker. Nobody else has tens of billions of execution traces from Perplexity, Groq, and Hugging Face's massive RL training rollouts for DeepSeek-R1 replicas.
Open-source clones can copy a 150ms VM boot. They cannot copy a taxonomy of 2 billion agent runs that predicts exact CPU, memory, and tool needs before the agent even executes line one.
To some extent, Perplexity's SPACE is the real competitor.
That's why E2B just released E2B Embed, literally handing customers a single-binary runtime to embed locally, while Daytona panicked and went closed-source.
Vasek is willingly commoditizing the "box" to monopolize the OS above it: Secrets proxies, MCP gateways, $36k+/yr Enterprise BYOC, and a self-driving control plane.
"Sandboxes are dead" isn't an obituary. It's E2B blowing up the bridge behind them so the copycats get stuck in 2024.
Deep Insight Labs retweeted
Happy JevDay!
But before announcements we have to talk philosophy…
The needs of early adopters / tinkerers / dreamers are quite different from what is needed for extremely stable infrastructure.
For the Jev release, we erred on the latter, but this release is for the former - with a gate so people can opt-in to the danger zone!
Deep Insight Labs retweeted
Saw this on Rednote and found it surprising (I am not able to verify it). It claims Baseten's gross margin was 17% in its most recent quarter, with the full-year guide cut from 30% to 20%. The stated reason: demand outran contracted capacity, so they're buying expensive short-term compute to keep customers served. It also says Cursor went from 57% of Baseten's annualized revenue in April to 28% in September, and that Baseten makes only ~4% gross margin on it.
I knew inference margins were well below SaaS, and way below data startups (I covered this in my earlier post on NVIDIA and open-weight models). But the public number reported for Fireworks is ~50%. Baseten doesn't disclose its own. 17% would be a very different business.
These numbers don't match what I'd learned about inference economics. But there's supporting evidence at another company: OpenAI reportedly told investors its adjusted gross margin fell from 40% to 33% in 2025, because it had to buy pricier compute at the last minute. I couldn't find a public source for the Baseten numbers. The closest thing: CEO Tuhin Srivastava said in June that a cloud provider quoted him $5.10/hr to renew a B200 cluster in October, up from $2.63. A 94% jump.
It's getting harder to tell which number to believe, and harder still to see where this is going.
Deep Insight Labs retweeted
Build a tax return software I dare you
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
Deep Insight Labs retweeted
congratulations team @reflection_ai
☀️ future is bright with western open frontier models ☀️
Introducing Beam: a highly efficient agentic open model with 501B total parameters and 23B active.
- Frontier reasoning efficiency
- Advances the Western open frontier on coding & agentic tasks
- Trained end-to-end from scratch
Full weights release this month.
Learn more about Beam: reflection.ai/beam
Deep Insight Labs retweeted
I tend to stay out of public fights like the one between @factoryai and @Cognition, because nobody walks out looking better. But, this one sits close enough to my own history that I want to add some context.
When I ran worldwide sales at @Nutanix, Frank Slootman was one of our advisers, and Mike Scarpelli, then @ServiceNow's CFO, chaired our audit committee. Both later went to @Snowflake, where they built the company alongside @CwDegnan. As CEO of @ThoughtSpot, we worked closely with Snowflake and with Chris. So I know Chris and the people who shaped him.
Scarpelli remains one of the highest-integrity operators I have worked with, especially on sales. That matters, because global sales hands you real ethical tests. I had my share at Nutanix, from FCPA exposure overseas to two-tier resellers who cut corners. He NEVER let me slide on even a simple issue. He was the CFO for Chris!
Three things are getting lost in the noise.
First, Chris was a board observer and adviser, and neither role carries a director's seat or a vote. A founder as sharp as Matan brings someone like Chris in for hiring, closing deals, and an enterprise network very few people have. That was a smart call.
Second, I know @Scottwu46 a little. I can't speak to every conversation. At Cognition's scale, though, a CEO has almost no time to steer by competitors. You track wins and losses, and you build the roadmap from what customers need.
Third, in AI a plan has a shelf life of about a month. Sometimes I don't know what @Tiny_Fish's plan will be next month😀. Anyone who leaves us carries a snapshot of a company that has already moved on.
None of this proves anything either way. Someone with access could misuse it, and the only real safeguard is trust in the integrity of the people in your orbit.
@matanSF is one of the smartest young founders I have shared a stage with, and he is building at remarkable speed. This won't be the last time someone close to him, or to any of us, joins a competitor.
All of this will pass, and these two companies will keep building for a while.
Deep Insight Labs retweeted
Have done a number of technical demos in the past and all I can say is
- give them a break, shit happens and things go wrong sometimes, even if you've practised a ton of times. might have been their screwup, the screwup of the team behind them, or just bad luck with wifi or whatever
- for those who are saying that they aren't prepared - I'd like to see you try and do better (you probably won't) - don't just be keyboard warriors
- they did an amazing job at recovery, props to that!
It's actually hard to be technical, eloquent and just ballsy enough to get yourself on stage in front of lots of people to do a technical demo. Give them some credit for that!
Two things went wrong during the @OpenAI #devday demos:
• During the dot demo, the dot wasn’t responding.
• In another demo where Codex was being used, voice chat wasn’t working at all.
And now I can’t even find these clips on the official stream anymore. 👀
But I recorded them live while the stream was happening.
Deep Insight Labs retweeted
Did not see this coming!
Synopsys CEO Sassine Ghazi says the deal only works because OpenAI is spending hundreds of millions to post-train a GPT Synopsys, while Synopsys keeps its IP out of any base model
"The requirement we have is our knowledge and IP cannot get sucked into a model that becomes the base model and, without our control, is available to the world. That's a non-starter for Synopsys."
"So the GPT Synopsys was a big investment from OpenAI. OpenAI will have to invest hundreds of millions of dollars to post-train GPT to make a GPT Synopsys. That does not come for free, so that's a big investment they have to make."
"The investment we are making is bringing skills, assets, R&D to work with them to fine-tune that model and make it achieve the best outcome between the intelligence and reasoning with our tools."













































