That's an editorialized headline. What they actually wrote was that it could be used to "generate misleading news articles, impersonate others online, automate the production of abusive or faked content to post on social media, [or] automate the production of spam/phishing content" and that they are aware other researchers have the ability to reproduce and open source their results, but this would give the community some time to decide how to proceed.
A nice game, though the truth table lighting round is pretty punishing! Big contrast to the circuit building part where you can take your time. Personally I'd drop the time requirements from that quiz section.
Having spent several years working on OT I can tell you that most of the gaps are things that should be waived anyway. Getting waiver files reliably integrated into that flow has been problematic as those files are fragile, alter the RTL and they typically break as they refer to things by line number or expect a particular expression to be identical to when you did a waiver for it.
This has all been examined and the holes have been deemed unconcerning, yes ideally there'd be full waivers documenting this but as with any real life engineering project you can't do everything perfectly! There is internal documentation explaining the rationale for why the holes aren't a problem but it's not public.
> Getting waiver files reliably integrated into that flow has been problematic as those files are fragile, alter the RTL and they typically break as they refer to things by line number or expect a particular expression to be identical to when you did a waiver for it.
Yeah last time I did this we used regexes but I really don't like that solution. I think the waiver should go in the RTL itself. I don't know why nobody does that - it's standard practice in software. SV even supports attributes exactly for this sort of thing. The tools don't support it but you could make a tool to parse the files and convert it to TCL. I've done something like that using the Rust sv-parser crate before. Tedious but not impossible.
Also we found the formal waiver analysis tools to be very effective for waiving unreachable code, in case you aren't using those.
> Also we found the formal waiver analysis tools to be very effective for waiving unreachable code, in case you aren't using those.
Yes we had used them just never got it slickly integrated into the verification dashboard. We had used this kind of analysis for internal sign off. You could generate the waivers manually and check them in but that suffers from the problem discussed above. Plus as OpenTitan was a cross company project you run into EDA licensing issues where not everyone has access to the same set of tools and a UNR flow could be running fine on one partner's infrastructure but isn't workable everywhere for multitude of reasons.
The ideal would be the nightly regression would do the UNR flow to generate the waivers and apply them when generating coverage but as ever there's only so much engineering time to go around and always other priorities.
I worked on OpenTitan for around 5 years at lowRISC. It certainly has its ups and downs but it's generated some great stuff and I'm very glad to see hit proper volume production like this. Whilst there's definitely open source chips out there and lots more using bits of open source that don't actually advertise this fact I believe this is the first chip with completely open RTL that's in a major production volume use case.
One of highlights working on OpenTitan was the amount of interest we got from the academic community. Work they did could actually get factored into the first generation silicon making it stronger. Ordinarily chips like that are kept deeply under wraps and the first time the wider security community can take a look at them development has long completed so anything they might find could only effect generation 2 or 3 of the device.
Academic collaboration also helped get ahead in post quantum crypto. This first generation chip has limited capabilities there but thanks to multiple academics using the design as a base for their own PQC work there was lots to draw on for future designs.
I'm no longer at lowRISC so I don't know where OpenTitan is going next but I look forward to finding out.
This is an interesting piece of hardware though when they go multi-chip for larger models the speed will no doubt suffer.
They'll also be severely limited on context length as it needs to sit in SRAM. Looks like the current one tops out at 6144 tokens which I presume is a whole chips worth. You'd also have to dedicate a chip to a whole user as there's likely only enough SRAM for one user's worth of context. I wonder how much time it takes them to swap users in/out? I wouldn't be surprised if this chip is severely underutilized (can't use it all when running decode as you have to run token by token with one users and then idle time as you swap users in/out).
Maybe a more realistic deployment would have chips for linear layers and chips for attention? You could batch users through the shared weight chips and then provision more or less attention chips as you want which would be per user (or shared amongst a small group 2-4 users).
> If copying user tokens was the bottle neck, batching would not achieve any speed up.
Reality is more complex. As context length grows your KV cache becomes large and will begin to dominate your total FLOPs (and hence bytes loaded). The issue with KV cache is you cannot batch it because only one user can use it, unlike static layer weights where you can reuse them across multiple users.
Emerging sparse attention techniques can greatly relieve this issue though the extent to which frontier labs deploy them is uncertain. Deepseek v3.2 uses sparse attention though I don't know off hand how much this reduces KV cache FLOPs and associated memory bandwidth.
> The issue with KV cache is you cannot batch it because only one user can use it
This is not really correct given how input token caching works and the reality of subagent workloads. You could launch many parallel subagents sharing some portion of their input tokens and use batching for that task.
1. Parallel investigation : the payoff form that is relatively small - starting K subagents assumes you have K independent avenues of investigation - and quite often that is not true. Somewhat similar to next-turn prediction using a speculative model - works well enough for 1 or 2 turns, but fails after.
2. Input caching is pretty much fixes prefill - not decode. And if you look at frontier models - for example open-weight models that can do reasoning - you are looking at longer and longer reasoning chains for heavy tool-using models. And reasoning chains will diverge very vey quickly even from the same input assuming a non-0 temp.
Anyone else find reading things like this slightly exhausting?
I'm very much pro AI for coding there are clearly significant capabilities there but I'm still getting my head around how to best utilise it.
Posts like these make it sound like ruthlessly optimizing your workflow letting no possible efficiency go every single day is the only way to work now. This has always been possible and generally not a good idea to focus on exclusively. There's always been processes to optimise and automate and always a balance as to which to pursue.
Personally I am incorporating AI into my daily work but not getting too bogged down by it. I read about some of the latest ideas and techniques and choose carefully which I employ. Sometimes I'll try and AI workflow and then abandon it. I recently connected Claude up to draw.io with an MCP, it had some good capabilities but for the specific task I wanted it wasn't really getting it so doing it manually was the better choice to achieve what I wanted in good time.
The models themselves and coding harnesses are also evolving quickly complex workflows people may put together can quickly become pointless.
I know people like this, theres a form of procrastination where they are busy hyperoptimising their todo lists and workflows but getting only a tiny amount of actual work done. It's a form of the disconnected intellect - they can tell you the problem, they can even tell you the solution, but they can't turn that knowledge into action. Convincing themselves that utterly trivial inconveniences are so productivity and or psychologically harmful they can they can then rationalize all this "meta-work" while simultaneously claiming to be virtuous, when in reality it's usually insecurity in their abilities or cowardice to face the unknown, preventing them doing real work
I almost agree with you, but with regard to this specific blog I'm not sure I can.
From my perspective, all this energy spent on AI prompting is actually just planning meetings and whiteboarding in disguise, but since all that has the bad reputation of luring devs into power struggles and yak shaving this is the new way.
It's likely where most of their improved productivity is coming from. The people doing the meta-work just need be mature about it to avoid procrastinating.
Yeah, AI generated diagrams can be pretty hit or miss. The lack of good quality examples in training data and minimal documentation for these tools can make it difficult for models to even get basic syntax correct for more complex diagrams.
I’ve had a lot of success dogfooding my own product, the Mermaid Studio plugin for JetBrains IDEs (https://mermaidstudio.dev).
It combines the deep semantic code intelligence of an IDE with a suite of integrated MCP tools that your preferred agent can plug into for static analysis, up to date syntax, etc.
I basically tell Claude Code to run the generated diagram through the analysis tool, fix issues it detects and repeat until fixed. Then generate a png or svg for a visual inspection before finalizing the diagram.
Now all of my planning and architecture docs are filled with illustrative flowcharts, sequence diagrams, and occasionally block diagrams for workshopping proposed UI layouts
Only taken a quick skim but this looks like solid material!
RISC-V Vector is definitely tricky to get a handle on, especially if you just read the architecture documentation (which is to be expected really, good specification for an architecture isn't compatible with a useful beginners guide). I found I needed to look at some presentations given by various members of the vector working group to get a good grasp of the principles.
There's been precious little material beyond the specification and some now slightly old slide decks so this is a great contribution.
The specification for an architecture is meant to be useful to anyone writing assembly, not just to people implementing the spec. Case in point x86 manuals aren't meant for Intel, they're meant for Intel's customers.
There is a lot of cope re the fact RISC-V's spec is particularly hard to use for writing assembly or understanding the software model.
If the spec isn't a 'manual' then where's the manual? If there's just no manual then that's a deficiency. If we only have 'tutorial's that's bad as well, a manual is a good reference for an experienced user, and approachable to a slightly aware beginner (or a fresh beginner with experience in other arch's); a tutorial is too verbose to be useful as a regular reference.
Either the spec should have read (and still could read) more like a useful manual, or a useful manual needs to be provided.
If you were an employee, you were not a founder. A founding-employee would be someone who explicitly "invested" time/money into a company without compensation. If you are also an employee earning a wage you better have a written agreement stating what amount was "investment" and what amount was compensated wage.
Startups typically offer employees, particularly early employees, substantial equity compensation. If the employer is offering this compensation in bad faith, or otherwise preferring one equity holder over another without an explicit contract - then they are at the very least a crappy business partner. A founding engineer with a 2% stake could be missing out on 5-10 million of this transaction.
As an aside, most founders are paid during the entire project. It’s not hard to raise a preseed round to get yourself paid for 6-24 months to work on an idea. If a founder chose to bootstrap - that’s all fine, but let’s not pretend that the employees aren’t taking massive career risks vs “standard” employers.
> If the employer is offering this compensation in bad faith, or otherwise preferring one equity holder over another without an explicit contract - then they are at the very least a crappy business partner.
I don’t know about you, but every company I’ve ever worked at is a shitty business partner if that’s the metric. The standard has always been I get what we agreed to if I was lucky, and otherwise I got full “I’ve altered the deal, pray I don’t alter further” and dared you to defend your rights.
I actually have called their bluff a few times and gotten some money out of it, but it was always a year long event or more to resolution and involved risking even more money on lawyers.
Just one slight problem: people need to eat, and food costs money.
Your startup won't succeed when its founder starves to death. It's why the founder will usually get a bunch of cash during investment rounds [0]: they can't focus on the company if they are constantly worried about cash in their personal life. Unless the founder is already independently wealthy, it is a guarantee that they'll be employed by the company and being paid a living wage. Heck, in some countries this is even legally required!
According to your logic, no successful startup will ever have a founder, as any form of pay instantly degrades them to regular employee, and any kind risk taken and below-market salary is completely irrelevant. Never mind the fact that they are taking home a minimum-wage salary while working 100 hours a week - they are earning a wage so they can't possibly be a founder.
So if this logic already breaks down for the founder, why couldn't it also break down for early employees whose compensation is mostly in stock options? How is their situation any different from the founder's?
Startups often pay a shitty salary in exchange for a decent chunk of stock options, with the implicit promise that you'll make bank if you work hard and make the company successful.
Getting screwed out of your payout by such a totally-not-an-acquisition is wage theft. It's like promising a sales-related bonus at the beginning of the year, and then in December changing the metric to "AI-related sales to the CEO's golf buddies".
Startup options are worthless. The only value most people will ever extract from a startup is the experience they had working there, and the salary that was put in their bank account.
I understand that a lot of inexperienced people (like in this thread) think they're going to get rich though.
No, it is not "wage theft" to not get rich when the company exits (by whatever means).
Nobody expects to get rich off working for a startup. The risks are massive, and very few exit with billion-dollar deals. This is taken into account by the people who work there and accept those stock options: 99.xx% chance of being worth essentially zero, but a tiny yet nonzero chance of being able to retire early when it does a billion-dollar exit. It's a lottery ticket, not a promise - every startup employee understands that.
Groq is now changing the deal after the fact by making those stock options worthless 100% of the time. It's like you participate in a lottery, and then the organizer decides to just not do a draw and keep all the proceeds for themselves. Sorry, but that's theft.
Don't intend to pay out in the unlikely event that you hit it big? Then don't offer stock options to your employees and pay market-rate salaries - plus of course a decent premium for the fact that (unlike an established company) your startup can go bust at any time and doesn't offer stable employment. You can't have it both ways.
Startup options are usually worthless, yes, because very few startups end up getting to a position where the options are worth something.
> No, it is not "wage theft" to not get rich when the company exits
I don't think anyone in this thread thinks they're gonna get rich by working for a startup. There's a hope that they will, that's why they are working, but there's no expectation. Maybe there's an expectation of getting a nice tidy sum after an exit (in the 5 or 6 figures) but not in the 7 or 8 figures, at least not if they're just employees and not founders.
What's being discussed is a startup exiting for billions of dollars and the employees with equity seeing zero of it.
Working for a startup usually means lower wages and longer hours, for the chance at striking it rich if the company succeeds. If employees don't see anything when the company succeeds, there's literally no upside to working for a startup.
I recall having to sit through many trainings on how to value employee equity. My experience is that most startup employers try to BS what it means to convince people to value their equity at a significantly higher price than they otherwise should.
If the employer is explicitly making the employee options worthless, then they should be obligated to disclose this. Otherwise it’s trivial to engineer a corporate entity which pays the employees while “licensing” the technology from an IP holding firm. Later they can simply sell the IP holding firm without owing employees a dollar.
The implicit promise is only partially true. Very rarely you can find a proven talent that will actually forego significant salary. Often time when that happens the person is close to founders and will have a significant role in shaping the startup and will get quasi-acquired too.
This promise may have been more true before 2010s where public companies were not paying as much in liquid cash and private companies were not valued so aggressively. Fact is most employees take the startup offer because they don't actually have a liquid offer that's super competitive at that moment, or they are just kind of bored and taking a break of the corporate job that does not give them too many responsibilities, i.e. they are compensated via the title, not just the promise of making bank.
That just means you’re pulling from the lower end of the talent pool. There is nothing wrong with this, but usually talent is correlated with outcome. Most hot startups which are going places are near impossible to get into even for folks with good offers.
Your last sentence is not mutually exclusive with my statement. Both could be simultaneously true. The sheer number of big company employees compared to hot startup makes it hard for everyone good to get into the startup, especially considering they usually have more specific needs. That said it could also be the case that the hot startup cannot easily get good employees from big company.
My point was more that the high end ones they do get are usually in the front piece of the airplane in the acquisition split. Also, the really hot startups are actually paying quite a bit of cash upfront so the original premise of employee sacrifice isn’t as true.
If you were an employee, you were not a founder. A founding-employee would be someone who explicitly "invested" uncompensated time/money into a company without compensation and also worked as an employee. If you are also an employee earning a wage you better have a written agreement stating what amount was "investment" and what amount was compensated wage.
In my career I've seen startups "shut down" and lay off the NA team.
I've seen venture capital acquire startups for essentially nothing laying off the entire product team aside from one DevOps engineer to keep everything running.
I've seen startups go public and have their shares plummet to zero before the rank-and-file employees could sell any shares (but of course the executives were able to cash out immediately).
I've seen startups acquired for essentially nothing from the lead investor.
In none of these scenarios did any of the Engineers receive anything for their shares.
Yet every day people negotiate comp where shares are valued as anything more than funny money.
As this gets more common, I think it will eventually lead to startups having a hard time attracting talent with lucrative equity compensation. It will be interesting to see how long it takes until this catches on among employees, but I already wouldn't take any positions in startups with a significant payment in equity anymore. The chances are slim that this pays out anyways, but now even when you are successful, noone will stop some megacorp from just buying the product and key employees and leaving everyone else with their stake in the dust.
At my last job search I didn’t consider any equity based startups seriously because of this trend. It was already such a tenuous path as it stood, but now with the norm established it seems like it’s become impossible for a rank and file employee to get paid out.
I’m more curious how angel investors are being treated in these exits. If _they_ dry up the whole pipeline goes away
Investors with enough into the deal to fight it in court get enough to not fight it. Key employees needed by the 'not acquirer' get compensation sufficient to retain them, although increasingly much of this is under a deferred vesting arrangement to ensure they stay with the 'not acquirer'.
Non-essential employees and small investors without the incentive or pockets to fund a legal fight get offered as little as possible. This structure also provides lots of flexibility to the 'not acquirer' when it comes to paying off existing debts, leases, contracts, etc. Basically, this is the end of being an early employee or small angel investor potentially resulting in a lucrative payoff. You have to remain central and 'key' all the way through the 'not acquisition'. I expect smaller early stage investors will start demanding special terms to guarantee a certain level of payout in a 'not acquisition'. I also expect this to create some very unfortunate situations because an asset sale (as they used to be done), could be a useful and appropriate mechanism to preserve the products and some jobs of a failing (but not yet fully failed) company - which was better for customers and some employees than a complete smoking crater.
that is a great point. it’s one thing to occasionally rugpull employees, who are still at least paid for their services and robbed only of their EV on their options (i say “only”, though i find this increasingly common practice to be absolutely deplorable, to be clear). but how could investors possibly be happy with this becoming the new normal? will it get to the point where these sorts of faux acquisitions also involve paying out investors and only shafting employees? at that point you are only really even getting like a 20% discount over acquiring the company outright, which hardly seems worth it. which is to say that your point is very astute: the investors are definitely the linchpin here.
The company still got $20B of cash(?) in its books, it can pay dividends to its shareholders (investors) and they get their payment. The company can go down the drain afterwards. If it can still make money with its remaining assets that's only a nice small bonus.
So the only ones getting shafted are the employees.
I suppose the firm could simply roll the 20 billion into a long term asset. It’s not a big deal to anyone except employees if the asset never pays out. Departed employees would not be privy to how the money is eventually exited from the now shell company 20 years hence.
true, but given that recruiters take upward of 30% of first-year comp it starts to really flirt with being a wash. if the acquiree’s employees are truly that odious, the founders aren’t worth the price of admission. just look at the scale.ai guy: he is in wayyy over his head now, and if those billions mattered at all to zuck he would probably regret it, ex post.
It's already happening. You need a good lawyer to read equity terms to make sure you aren't going to get rug pulled by a founder later on. Even so I still consider equity to generally be worth zero unless the founder is someone I trust fully, since there are so many ways for them to legally not give you anything.
The equity in almost all startups has already been a bait and switch for more than a decade. Most will refuse to answer you about % of equity share anyways, but if they did it's tiny tiny amounts, and in the end half the time it's up to the acquiring entity just how seriously they end up taking it. If you landed at an entity like Google (as I did from the place I was working 15+ years ago) you could be treated well. Elsewhere, not great.
During boom times it made more financial sense to go straight to a FAANG if you could.
If you ask me there has been a major shift into trying to make "startups" into just another form of corporation. It started years ago when I started seeing things like "Founder Engineer - 0.5% equity" in jobs here.
Can you say more about why mechanically she didn't get anything?
If you exercise your options you have real stock in the company, so I don't see how you can get shafted here.
Did investors do some sort of dividend cash out before employees were able to exercise their options? (Obviously shady, but more about investors/leadership being unethical than the deal structure).
Would love to know more about how this played out.
Multiple share classes are the norm even before the new acquisition types we see here. It’s extremely common in an acquisition for employee shares to be worth nothing while investor and founder shares are paid out.
But these new “acquisitions” aren’t even that. They are not acquisitions at all. They just hire the talent directly with perhaps an ip rights agreement thrown in as a fig leaf.
I'm well aware of dual class shares, but preferences are typically 1x, and none of the deals were for less than the amount raised, so they're not relevant here.
The fact that these are not really acquisitions doesn't change the fact that Groq the entity now has $20b.
Money can't just "go" somewhere, it needs a reason first, at least for book-keeping. I mean, VCs can get their invested capital back but on top of that, how would that money be transfered? $20B is a lot and for sure the VCs will not just write an invoice of $18B for consulting services.
Hey, husband of that friend here,
The bought company had huge debts to the investors (it is a startup, not tiny but small one, ran for several years) and after cashing those out from the purchase deal, the employees were left with shares that were worth 0$.
(might be that the founders also grabbed some money out of that purchase, no one knows tho)
The employees of that bought company were given an incentive by the buying company to stay for a while and help tearing down and integrating their product into the buying company.
One could say shady, I'd say that it was just a bad deal.
It's definitely true that common stock gets $0 if the acquisition price is <= (sum raised + debt).
That sort of sounds like the startup wasn't doing well, and the acquisition wasn't for a lot of money (relative to amount raised), which seems very different from these Groq/Windsurf situations.
There have been at least a half dozen of these deals in the past 1-2 years including Google “licensing” CharacterAI to pull their founders back into Google as valued employees.
In the deal mentioned above: my guess is that preferred class shareholders and common shares got paid out but the common shareholders had such a low payout that it rounded down to zero for most employees.
This can happen even in a regular acquisition because of the equity capital stack of who gets paid first. Investors typically require a 1x liquidation preference (they get their investment back first no matter what).
Liquidation preferences are typically 1x these days, so they only matter when companies are sold at fire sale prices where basically nobody is making any money.
The deals are all weird so it's hard to really know what's happening, but if Groq gets $20b, I don't see how common stock holders don't get paid.
Special dividend to priority class and retain the rest to grow the remaining sham company?
I've seen some discussion that paying out normal employees might look more like an acquisition on paper which they may want to avoid for ftc reasons. I've also seen some discussion that this is a quid pro quo to the trump family to get Nvidia back into China (jr. bought in at the September financing round..).
Lots of speculation in general, including why nvda chose to spend 20bil on this.
I was involved in a (obviously smaller) situation with an acquisition that went to a top consumer CPU maker (you can guess). The investors got nothing as the buyout money was used to fund new pivots in the existing company. So no options or shares were monetized and investors maintained their existing stake that had technically the save value, just most of the value was temporarily all cash. The only people to make out were the ones who went with the asset sale (retention bonus stuff) and the leadership that stayed (raises, etc.)
reply