> They didn't have to join, which means they got a solid valuation.
This isn't really true. It's more about who wanted them to join. Maybe it was Anthropic who really wanted to take over Bun/hire Jarred, or it was Jarred who got sick of Bun and wanted to work on AI.
I don't really know any details about this acquisition, and I assume it's the former, but acquihires are also done for other reasons than "it was the only way".
The year was indeed 2015, and his political content would fit OP's description, but the reason for the ban was that he was repeatedly being a total asshole, not "YC is scared of anti-fascists".
You know... one of the charges leveled at Michael was creation of sockpuppet accounts in various media (Wikipedia, HN, Quora). I suppose it could be within the realm of possibility that this is another one of these, given the oddly specific example in the middle of general musings. The fact that all follow-on replies have to do with this particular incident and nothing else in the post, painting the victim in the most positive light, makes this especially questionable. If I'm wrong, I apologize and wish the OP well, but I gotta say, it looks pretty sus to this observer.
How can you justify such a wildly conclusive statement without providing any supporting information? It's not just the "HN Bubble", recent articles on The Register discuss how there is no way to guarantee the information provided to the 3rd party LLM servers in the prompting is free from further disclosure. This raises a host of concerns for professionals with a non-deligable duty to safeguard client/patient information from disclosure. To the point that if LLM captures that information, by definition the attorney or physician has disclosed client/patient information putting their license to practice law or medicine at risk.
Exacerbating the matter is there is zero disclosure on Mozilla's part detailing exactly what information is sent to 3rd party servers as part of its AI rollout in FF 145. Would you risk your license on some FOMO AI rollout in FF that, unless you, as the lawyer or doctor, have stepped though each line of the tens of thousands of lines of FF code associated with this new AI to meet your ethical obligation and answer the bar or medical board's inquiry on whether client/patient information has been sent to a 3rd party?
Without the ability to completely disable and turn all AI submissions off, Mozilla's "Trust Us" position doesn't allow anyone with such a duty to meet it. This is before you even get to confidential and proprietary or trade-secret type information applicable in any professional setting.
These are all vexing questions from a legal standpoint.
By saying he was part of these conversations, the people he talked to almost certainly already know who he is (if necessary, the fact he's undergoing through surgery definitely tells them). Not sharing the details about this sounds more like he doesn't want this individual to be rehabilitated. (Assuming this entire thing is not made up.)
You can agree with the goals of an initiative and still think it is poorly thought out.
OP is acting as if anyone criticizing this thing must clearly be opposed to their entire world view, accusing them of being paid shills. No. Maybe they just (rightfully) don't like Clippy, and don't want a movement they care about to turn into that.
How would a human classify the cancers? I assume the LLM training data does not include a whole bunch of cancer samples, so assumably there are some rules that it follows?
> While there exists several pathology-focused AI models
Would also be curious how the LLM compares to this and other approaches. What's the performance of the models trained specifically on this task, and random guessing, compared to the expert pathologist? Correct me if I'm wrong but this seems like the sort of task where being right 90% of the time is not good enough, so even if the LLM beats other approaches, it still needs to close the gap to human performance.
> What's the performance of the models trained specifically on this task, and random guessing, compared to the expert pathologist?
I should probably first clarify here, the disease classification tasks are about subtyping the type of cancer (i.e classifying a case as invasive ductal carcinoma of the breast) rather than just binary malignant/benign classification so random guessing is much more difficult and makes this model performance more impressive.
> Would also be curious how the LLM compares to this and other approaches.
There aren't a lot of public general purpose pathology benchmarks. There are some like (https://github.com/sinai-computational-pathology/SSL_tile_be...) but focus on just binary benign/malignant classification tasks and binary biomarker detection tasks.
I am currently working on self-hosting the available open-source models.
> this seems like the sort of task where being right 90% of the time is not good enough, so even if the LLM beats other approaches, it still needs to close the gap to human performance
Yep, your intuition is right here, and actually the expectation is probably closer to mid-high 90%, especially for FDA approval (and most AI tools position as co-pilots at the moment). There is obviously a long way to go, but what I find about interesting about this approach is that it allows LLMs to generalize across (1) a variety of tissue types and (2) pathology tasks such as IHC H-score scoring.
I don't get this logic. Putting aside that to get 33 different models you would come up with 5-6 different form factors, each of them on a distinct point in the tradeoff scale, why do you think that something is only worth doing if it can be put on an uniform supply-demand curve?
Apple made nearly $190 billion last year selling just iPhones.
If you think it costs more than $5 billion to design a phone and set up a production line, you are wildly off base. That’s the kind of money companies spend to build silicon fabs or release half a dozen new car models, not consumer products made by a contract manufacturer.
...you do know that Apple produces its own silicon, and probably uses about an entire TSMC fab's worth of capacity? In the end, the money to build that fab is coming from Apple.
Apple isn't making average consumer products with average contract manufacturers.
If Apple didn't run such a closed ecosystem, other hardware vendors would step in and be happy to sell a form factor that 3% of the market uses.
I keep trying to use Andriod to get more choice on form factor, but one thing always brings me back to an iPhone: texting incompatibility. Apple has me locked into their ecosystem because I can't get a decent quality video texted to me.
As an Apple fan since the 90s who remembers how Microsoft abused its market dominance for decades, it's particularly ironic that Apple continues to use this technique against other companies.
There is a wide variety of form factors available in the android ecosystem. Whether or not they fit your definition of "decent" just depends on how much you prioritize size:
Foldables get this job done well. My (OG) Pixel Fold is a great size & aspect ratio while folded, easy to use one-handed, but has a giant screen when you open it up. The newer Pixel Folds and the other foldables on the market have all grown the screen vertically but they're still more compact than most flagships.
I wouldn't. I personally think iOS kind of sucks, and I only keep using it because Android developers don't support devices long enough for me. Third party developers would be as much a mess as they are in the Android world and at that point I'd rather have a phone with a good OS.
> I can't get a decent quality video texted to me.
It seems this gap has significantly closed, assuming both sides have RCS support. I've got a number of decent quality videos sent through RCS from friends through RCS.
Apple developed iMessage to work around the problems with SMS and MMS, as well as decrease load on carrier networks. There is no closed ecosystem, you can still receive messages and videos from iPhone users, just at the quality your hardware and software can support.
Google later decided to come up with a completely different implementation called RCS to deal with the same problems. Rather than work with Apple on bringing an iMessage app to Android or licensing it, they instead tried to pressure Apple with a public advertising campaign to adopt what is frankly an inferior solution that doesn't even have reliable end-to-end encryption.
Your complaint is basically that you bought a Toyota and it does not have BMW's laser headlights that adjust brightness and angle automatically. You still have headlights, you just didn't spend the money to get the good ones.
Yes because Google offered nothing of value in return. Like in my example, nothing stops Toyota from offering enough money to license BMW's laser headlights.
Google tried the same thing Apple did long before RCS when it made Hangouts the default SMS app for Android. Conversations could be upgraded from SMS to Google's internet-based chat protocol if the other person had an account; it was even available for iPhones, but it couldn't be an SMS client.
Lets be honest, Hangouts (which of the three versions) was a crappy chat app that they wanted to boost the usage of. It wasn't intended to be a functional SMS replacement.
> just at the quality your hardware and software can support.
I assure you, Android phones have been able to render video of far higher quality than what Apple devices would send to them through MMS.
iMessage is a cloud ecosystem. I cannot install iMessage on my Android device.
> Rather than work with Apple on bringing an iMessage app to Android
Apple has been free to release this app at any time. There is nothing Google is doing that prevents it from being made. The only people preventing this app from existing are the people at Apple.
> Rather than work with Apple on bringing an iMessage app to Android or licensing it
This seems like an unfair take - Apple is on record using iMessage specifically to deteriorate the experience between Android and iOS users. I don't see them working with Google to bring iMessage to Android.
Not to mention the SE (variant of the 4 I think?) was way more popular. Dismissing the whole concept just because one implementation at one time was a relative flop (and as you point out, that's still a lot of sales).
Also, they're happy to have Pro and non-Pro SKUs etc., just averse to smaller for reason.
It's a self-fulfilling prophecy to launch the iPhone mini during Covid, offer it 2 years, and say no one wants them.
The haters dismiss the SE point by saying it was the price, not size. But it does prove that no one avoids phones due to being "too small".
The size increase is because of cost optimization. Where it's wrong is that everything else meant for humans comes in small, medium, and large. Phones? Just large and XL.
Good, in some ways. But do people want to pay higher prices for these iPhones to cover the costs that have to be amortized over a lower volume of devices?
This looks great. I'm probably gonna keep the threshold set to 0%, so a bit more gradient variety could be nice. Red-yellow-green maybe?
Also, can I use this on AI-generated code before creating a PR somehow? I find myself spending a lot of time reviewing Codex and Claude Code edits in my IDE.
Either would work, I think. How I do it right now is that I let AI edit automatically, but then check the diff in Cursor before I stage my Git changes. May be different for others.
Yeah, heatmapping the diff before creating a PR would need tighter IDE integration. We're working on cmux for this purpose. It's kinda an IDE, and it lives in the same repo: https://github.com/manaflow-ai/cmux.
We clearly don't have the same friends then! The event was the ridicule of the day on Twitter (wrongly, in my opinion, given that the tech is very good).
Hindsight bias, and you're wrong by saying that the $5k were conditional on the line item change (they weren't). Here's the quote from the indictment:
> JAVICE responded to Scientist-1 that same day, “Can you send the invoice back at $18k and just one line item for data analysis?”
>
> Scientist-1 replied, “Wow. Thank you. Here is the new invoice.” Attached to this email was a new invoice, now for $18,000, with the previous specific descriptions removed and replaced by the one-line description, “Data science services”
It is very common for large companies to ask for changes to line items as it helps the finance team categorize. Also, it's clear from this conversation that the scientist believed the $5k to be a generous tip.
Even if all of that weren't the case, this should be a case of innocence until proven guilty; it's easy to see the crime in hindsight, but if you suspect nothing that thought might just not cross your mind in the first place.
You can call it hindsight bias if you want. What you seem to be saying is that your heuristics would have told you that Javice had no nefarious intent based on that exchange, and you would of course have been wrong. My heuristics are very different.
It is very obvious to me that the extra $5k was "conditional" because if Scientist-1 hadn't sent back an invoice for $18,000, he wouldn't have gotten paid $18,000.
I have never heard of a company giving a 40% "tip" after the fact to a contractor who was already billing $600/hour. It seems like Scientist-1 did not question this tip because it didn't benefit him to do so. He didn't suspect anything because he was motivated against suspicion. That's a lack of integrity.
I don't really know any details about this acquisition, and I assume it's the former, but acquihires are also done for other reasons than "it was the only way".
reply