New paper out of the Max Planck Institute for Intelligent Systems. If this holds up, it seems big.
Abstract: The continued improvements in language model capability have unlocked their widespread use as drivers of autonomous agents, for example in coding or computer use applications. However, the core of these systems has not changed much since early instruction-tuned models like ChatGPT. Even advanced AI agents function on message exchange formats, successively exchanging messages with users, systems, with itself (i.e. chain-of-thought) and tools in a single stream of computation. This bottleneck to a single stream in chat models leads to a number of limitations: the agent cannot act (generate output) while reading, and in reverse, cannot react to new information while writing. Similarly, the agent cannot act while thinking and cannot think while reading or acting on information.
In this work, we show that models can be unblocked by switching from instruction-tuning for sequential message formats to instruction-tuning for multiple, parallel streams of computation, splitting each role into a separate stream. Every forward pass of the language model then simultaneously reads from multiple input streams and generates tokens in multiple output streams, all of which causally depend on earlier timesteps. We argue that this data-driven change remedies a number of usability limitations as outlined above, improves model efficiency through parallelization, improves model security through better separation of concerns and can further improve model monitorability.
"Hundreds of companies rely on Stainless to generate SDKs, CLIs, and MCP servers—the libraries, command-line tools, and connectors that let developers and agents use an API."
I'm waiting for the Enterprise space to wise up. For anyone who's ever worked with any reasonably large company as a vendor (especially a small one) you know how painful redlines in legal can be. Why TF haven't enterprise made it more painful for these events? Basically state that if the service is purchased/sold/shuttered prior to the contract expiry date that a significant penalty (e.g. full refund) is required and including some portion of investment made to onboard said service/product/tool.
I can't even imagine the money wasted on turn-and-burns in the F1000 alone. The US needs a wake up call with respect to consumer / buyer protections. The life of the snake oil salesman is plentiful these days, and you have a lot of AI-psychotic executives who can't seem to get enough.
> Why TF haven't enterprise made it more painful for these events?
They mostly have. By mostly refraining from dealing with startups and companies they deem either “too young” or "too small" to be reliable partners. And, when they do, imposing long sales cycles.
And thus the enterprise well is poisoned for most startups.
A place I worked some years ago we even had an escrow foisted on us by our larger partner in the agreement so that they’d be able to continue running the software we were building if we went under.
Honestly, it was a pain in the ass and meant that for them alone we ended up running an older version of the software than we offered to clients because as we developed its capabilities it became ever more integrated into our core platform and we weren’t about to escrow that.
When the agreement came up for renewal at the three year mark we managed to get the escrow clauses removed.
A lot of money is made this way. It'll take an act of Congress (or something on that level) but many of us are already "on the take" so to speak, so I doubt it'll ever happen.
This is why it's good to consider an open-source product backed by an enterprise support company. Growthbook is an example. If they go poof you still have dozens to hundreds of other companies, and open source base, and can collaborate with the other users (companies) to crate a foundation to carry on development if needed. Or just patch it yourself. There's a continuum depending on how critical and how deeply you exploit it.
what is the value in destroying those relationships? I assume it was acquisition to defend against another company owning a key part of their delivery pipeline, but killing the public product is just bad press.
the relationships and enterprise customers they have are probably wildly blown out of proportion and few if any actually used the product in production.
They can also keep the product running behind the scenes for a select few and just shut down the public facing part
What's WILD is people ending up relying on these essentially startup-slops that just serves to give you future technical debt once you have to eventually moved away because they got acquired by $INSERT_BAD_GUY_OF_THE_MONTH
The only people "relying" on this are other startups whose VC benefactors force them to use other products under their portfolio in order to goose up their numbers.
For context I was the founding member of our customer eng team, the main drive for Stainless adoption came from engineering teams who didn’t have the time/capacity/expertise to implement high quality SDKs + maintain the whole release pipeline automation across multiple languages. The pitch is „we want you to focus on your domain of expertises. We handle all the plumbing for your end users to have the best DX possible when using your services“. Which is IMHO very compelling
It is indeed, and I totally understand why people want something like that! Shame about the closing of the service though, which will pretty much make everyone turn away from similar solutions in the future if they're provided by a startup, as the rugpull Stainless is about to do will be a bitch to deal with for many.
The goal for the customers was to save time, little did they know Anthropic had other ideas for them :)
I see it differently. I don’t think it is fair to describe it as a rug pull, the sdks we generated are owned by customers and aren’t going away. They own the repositories. We also implemented a self-service option just for the transition out from Stainless SaaS.
The most important part in my opinion is that with Stainless we were able to prove that a profitable market for SDK codegen exists and that people do care a lot about the quality of the generated code. New and existing players can take advantage of the segment we contributed developing. We also proved that a whole set of companies are ready to pay to provide an excellent developer experience to their end users.
At the same time the whole ecosystem is being reinvented by AI, it is possible that Stainless as it was doesn’t make sense in a future where everything is agentic, it’s hard to say.
I personally believe there is still a lot of values that can be found in the space Stainless built — and I’m sure Fern and others will continue to grow and develop solutions around OpenAPI
You may not even see it. I worked in a startup whose founder had money dipped into about a dozen products in the cyber security vertical. Many of those startups, I later found out, had access or used products from others in his portfolio. Basically taking $50k and cycling it through all of them buying something from the other one. I doubt it was a money laundering scheme, but it sure was convenient to just add logos of "customers" to the Nascar pitch slide.
Go to the website of pretty much any AI startupslop, Google who led their series A, then Google who led the series A of the other AI startups (it’s always other AI startups) whose logos they show as users/testimonials/case studies on their landing page. You’ll start seeing a pattern.
that makes so much sense. I always wondered how the fuck did all those ZIRP era "hello world as a service" bullshit startups have any customers at all.
Stainless wasn’t created during the zero-interest era. And we had paying customers since literally the first month of existence. We developed everything in close collaboration with customers
It may be that there are many projects relying on Stainless, or, as a sibling comment points out, it may be portfolio-based stack selection rather than actual feature dependence.
Either way, it does seem irresponsible and tone deaf for an acquiring/hiring company and an acquired/hired company to send these conflicting signals. If one puts oneself out there as dependable in the face hopes and needs of other, smaller, up-and-coming projects, then a rapid wind-down for $ is incongruent with such a posture.
So much so that, at least for my part, I'd be quite reluctant to hire someone who had engaged in this sort of bob-and-weave pursuit.
Kinda, most hams are very rhadamanthine about following every tiny rule to the letter, or their even stricter interpretation of the spirit. The type of people who complain about young people not joining the hobby while insisting on maintaining strict licensing rules and tests. It's very much the polar opposite of hacker ethic.
The two biggest ham people I've known both, independently and separated by years, discussed enjoying war driving looking for "pirate" radio signals that they can report to the FCC. Amusing to find this is a cultural aspect of ham radio licensees.
I've got friends in the scene and their behavior about it reminds me of other types of friends I have
1. Amateur pilots
2. People late in their years getting into martials arts for the first time (will be the loudest "KIYA"s in the class and always doing the most aggressive deep bows lol)
3. Non libertarian gun nuts. Oh buddy the attention they pay to everything from how you load your gun at the range to how you've had it packed in your car.
I have specific individuals in mind for each of these categories and I say this without ill intent, I'm not trying to disparage this behavior, it just seems to be a specific kinda thing, where following the exact letter of written direction seems to be half the fun for them.
This in opposition to some other types I know who aren't having fun unless SOME rule is being broken...
This + amateur radio is designed to be open, no encryption, anyone can talk with anyone, and if you're "being stupid" (to not use other terms), anyone can tell you so.
The "secret, encrypted, private" chats correlate more with random "doomsday preppers", and younger non-hams (cheap, no need to get licenced). Many of those people buy (ham) radios too ("for emergencies", can't transmit legally anyway), but don't really contribute to anything. Emergencies are handled by trained groups of hams when/if they're called to help by whatever proper agency needs help with communications.
You can legally transmit for any emergency, regardless of licensing status. The emergency definition is purposefully left ambiguous so as to apply to many situations. Lost in the woods? Go ahead, transmit; the FCC isn't going to knock on your door if you live. Every HAM should know this already.
This is the correct info. Anyone in an emergency is allowed to use the amateur frequencies. Just got my technician license a couple weeks ago, emergency use is even actually on the test!
> No provision of these rules prevents the use by an amateur station of any means of radiocommunication at its disposal to provide essential communication needs in connection with the immediate safety of human life and immediate protection of property when normal communication systems are not available.
That rule applies only to amateur stations (it says so right in the text!), not unlicenced individuals. What an amateur station is is defined in the beggining of the document, and yes, that requires the a duly-authroized (licenced) operator.
The last thing you want in an emergency event is some prepper with a baofeng transmitting on a repeaters frequency without a subtone set (because he's too stupid to pass an exam that 10yo kids can pass) effectively jamming it for proper emergency users. The other thing is, that chances are no one will actually hear you, especially on simplex. With tools like garmin inreach, carrying an HT with you instead of something proper and relying on that to save you in a time of need is just stupid.
Ham radio is like driving, you need experience to do it, and even some experienced people still do it badly. Trying to figure out how to drive by reading a car manual while the flood waters are rising is going to be a pretty bad experience.
Well, according to multiple times where people have checked with FCC enforcement folks, the spirit of the ruling covers unlicensed users operating in amateur bands for real emergencies.
The rules are clear here, there is no "spirit" in the law. The problem is, that the myth of somehow being "saved" by having a baofeng with you is spreading and people will die because of that. Hopefully only the baofengers and not others, affected by people who would effectively jam multiple others operating on eg. a repeater.
Ham radios don't just appear, someone has to buy them, buying one without getting a proper licence is just stupid.... but many (especially americans) do so. There is GMRS, there is FRS, people could take those radios, try them out when not in an emergency, but nope... everyone wants that uv-5r for some reason.
Every one of those preppers should get licenced first, go to some hiking trail, some remote..ish pota park and try to do an unspotter POTA activation there... and after failing horribly, they'll rethink their emergency communications. Somehow even licenced hams (about which I assume none actually tried doing an unspotted pota from some hiking trail) support and spread the "just buy a baofeng for emergencies". In reality... they're useless in most cases. If you're somewhere remote, no one will hear you anyway, and if you're stuck at home, having something like a starlink will actually help you reach someone, much better than a radio, especially a handheld 2m/70cm one. You might get some good but useless DX with an HF one, but you won't be setting up an NVIS antenna in a snow storm.
the rules arent clear, which is why 97.403 and 97.405 have been argued nauseam for a VERY long time. It's intentionally vague so that some people cant claim being out of soap as an emergency. But during a real emergency (someone had a heart attack on the floor, you're being chased by an axe murderer, etc) every single representative has said that the spirit of those parts covers the person. I don't disagree that people should get licensed, mostly so they know how and where to operate the radios or god forbid we do have a collapse and do have to fix up our own radios and antennas.
It's not actually vague, if they wanted "everyone" to do that, they'd use "everyone","anyone" or some similar wording, not specifically limit the exception to amateur station. Someone chose that for a reason, and it's a good reason... even licenced hams can cause trouble (eg. forget to turn off a simplex repeater on a radio, "mars mod" it and then jam the fire department frequencies with it,... a few weeks ago), and people who can't pass the simple exam don't need such radios, they can get by with GMRS (if they're from US and able to fill out an online form) for FRS/PMR (if they're not) or even MURS (again in US). There is no real difference in range between an uv-5r and a uv-5g (ham and gmrs radios), and a very small difference with frs/pmr (you need a line of sight anyway).
If someone had a heart attack, somewhere remote enough, that there is no cell signal, using a GMRS radio will have the same effect and range than a ham radio (i'm talking about 5W HTs here). Using something like a garmin inreach would actually get them help, but still, preppers want their baofengs, and for some reason don't want the *g models. That's why i get bothered when people promote ham radios, especially baofengs for emergencies, because they'll be useless in most cases and people who don't know that, will rely on them instead of getting a proper tool for the job. Many of those even have that in their pockets right now (some samsungs and iphone can do satellite communications already). Promoting the "you don't need a licence in emergencies" and then turning to "you'll be breaking the law anyway but who cares" mentality means that people don't learn even the basics (if they did, they'd be able to pass the exam) but still rely on those radios to get help.. and in turn, people will die because of that.
As i said before.. if you have a heart attack in the middle of nowhere, a baofeng won't get you help. If you went with gmrs/frs, you'd test it out and see the limited range and that no one is actually listening out there (unless arranged, and that person is in simplex range), if you get licenced, then you'd do the same, call out cqs into the void until you got bored, but if you do the "just buy one, you don't need licences..." (even if you do need to be licenced), people will be using that radio for the first time during an active emergency and fail in getting help with them. Stop promoting the untrue myth of getting help with a ham radio, instead offer proper tools for the job and people will actually be able to get help.
You're arguing for why people shouldn't buy a ham for prepping without a license or weirdly arguing that someone who is totally unprepared should've just happened to have a garmin on them. Realistically most people aren't toally prepared. A wife went to a remote cabin with their husband who is a ham op and has a heart attack, someone falls off a cliff while taking a photo of someone with their phone and the only thing is a mobile vhf/uhf radio in their car, etc etc etc. there are countless reasons why someone unlicensed might end up needing to transmit on amateur bands in an emergency, which again is why it's vague. You're claiming its not even though FCC enforcers themselves have said otherwise, so you should probably go argue with the fcc instead of strangers on the internet.
you'd think with how often Opus builds two separate code paths without feature parity when you try to vibe code something complex, people wouldn't regard this whole thing so highly
LFP was developed in the late 1990's and NCM in the early 2000.
They have already gone through multiple iterations: NCM523 first was mass-produced in 2007; the latest mainstream NCM is now NCM811, followed by the next-gen NCM9.5.5 (higher density). LFP is now up to the 4th generation.
That being said, the EV batteries aren't just driven by improvements in the cathode, but also in the anodes, such as silicon composite, or in this particular case, anode-free batteries.
LFP are common in EV’s and ‘solar generator’ style battery packs, but I’ve never seen them in phones or laptops (outside OLPC), reduced capacity makes them not great in these, better to go NMC.
which "high-end" phones and laptops use LFP? This makes no sense.
Also, LFPs are mostly deployed in low-range EVs and mostly in China. Most EVs outside China still favor NCM/NCA, but I suspect that LFP is going to gain market share in budget friendly, low-range EVs.
In China, the current trend in "high-end" smartphones/laptop is to switch to higher-energy silicon-anode batteries. LFP is primarily for low-energy dense (gravimetric/volumetric) storage devices, such as power banks; or in vehicles, low-end/low-range EVs, or stationary energy storage (ESS) -- BYD being one notable exception.
China implemented rules in mid-2025 banning uncertified (no 3C mark) and recalled power banks from Anker & Romoss models, from domestic flights due to fire risks.
