Hacker Newsnew | past | comments | ask | show | jobs | submit | behnamoh's commentslogin

I don't think that's what this issue is talking about. I have the Max $200/mo plan and have noticed starting yesterday that my quota drains much much faster, to the point I'm about to use the $50 credit Anthropic gave away to everyone.

what kind of evidence would you expect? it's not like everyone has a claude code monitor with detailed logs.

Why is it on hacker news?

That’s the implied question. It’s an issue on a piece of software. There’s more than 10,000 of them.

Why is it news?


How does Tailscale make money? I really like their service but I'm worried about a rug pull in the future. Has anyone tried alternative FOSS solutions?

Also, sometimes it seems like I get rate limited on Tailscale. Has anyone had that experience? This usually happens with multiple SSH connections at the same time.


Our company pays for the premium business plan, $18/mo/user. You have to pay for at least the lower tier plan once your team grows beyond a handful of people. And there's several quite useful features (though maybe not essential) on the premium plan like serve/funnel and SSH.

On the other hand, I do wonder about zerotier. before tailscale we used zerotier for a few years, and during the first 3-4 years we paid nothing because as far as I can recall there was nothing extra that we needed that paying would've gotten us. Eventually we did upgrade to add more users, and it cost something like $5/mo (total, not per user).


Zerotier is not the same as tailscale although both can be used to do the same, but under the hood both are fundamentally different, ZT is layer2 like switch, so it’s like an Ethernet meanwhile TS is built on top of wireguard and is layer3. ZT allows broadcast/multicast and has own protocol, TS don’t. I use both among others, and ZT since around 2019, I found it reliable in some cases in IoT world while TS had better throughput in usual applications.

Yeah, they're not direct replacements. I think both models have have their pros and cons. In fact I tried both around when covid shutdowns started (server being in the office, me at home), and liked zerotier better; it was faster, and a more generous free tier. But now tailscale has won out for a couple of reasons; the main one, it's simply less flaky for us on macOS, especially for devs working overseas. No idea why and maybe there's simple fixes (that don't involve repeated connections/disconnections, hopefully). The other, tailscale has a few extra things that are nicer and easy to use like identity-based ACLs, funnel/serve, magicDNS, ssh management, etc.

Zerotier works fine for me, but with a huge exception which I just can't figure out. On my Linux laptop which also runs OpenVPN and with some specific routing set up, Zerotier will, after a minute or so, completely take over the routing and default everything through Zerotier, and nothing I do with the routing tables will change this. I have to stop ZT at this point and then it reverts to normal. Every other computer in my ZT network behaves fine.

This is so problematic that I'm considering looking into Tailscale, I understand they work very differently but it looks like my use case could be covered by both.


I had to do MTU tuning on macos on the ZeroTier interface (find your feth name via ifconfig)

# Replace feth1234/feth2345 with your active interface

sudo ifconfig feth1234 mtu 1400

sudo ifconfig feth2345 mtu 1400

..and for working with Windows peers, manually "Orbit" the Windows Peer as well as adding a direct routing hint for the internal ZeroTier IP. ZT definitely takes some effort for tuning.


I've used serve/funnel on the tailscale free tier... definitely agree that the team size limit seems like it would move companies to the paid plan though.

I think how it works usually is that they let you use the features from higher tier plans than the one you're on; once you use them enough they send you an email asking to upgrade. That's what happened to us and I've seen other users mention it. Not sure how I felt about it, OTOH maybe it was less friction than explicitly subscribing for some "2 weeks free trial" or whatever but OTOH it did feel weird and unexpected. Anyway, we felt the extra features were worth it so ended up paying.

Hmm...

Ok I checked the pricing page and funnel is available in the free tier (limited to 3 users) but not the $6/user/month tier - which you need for more than 6 users... strange pricing structure but I guess I see the logic.

Any chance you were asked to upgrade from $6/user/month to $18/user/month and not free to $18/user/month?

https://tailscale.com/pricing#application-networking


Yes, we were on the starter $6 plan. The feature we got messaged about was SSH management, iirc.

How do you handle the do-before-thinking devs? Or the kinda low-to-mid performing devs? Most companies has one or a few of those, right? They help the company machine go around by doing the somewhat boring stuff over and over again.

Tailscale in a company/developer env seems awesome when you know what you are doing and (potentially) terrifying otherwise.

Does someone set up detailed ACLs for what's allowed? How well does that work?


> How do you handle the do-before-thinking devs?

Isn't that exactly what tailscale is built to accommodate - zero trust?

You set up ACLs and other permissions to not allow people to do more than the damage you can tolerate.


Zerconf ≠ zero trust. The difference could not be more material in this context.

If both sides of your ssh tunnel (pub,private keys) are under your control, in theory, that's "zero trust".

Unless one considers the meta data such as src/dest IP are visible to Tailscale sw.

Right?


'Zero trust' has a technical definition that's not really relevant here. See: https://en.wikipedia.org/wiki/Zero_trust.

The concept is separate from 'zero config' (https://en.wikipedia.org/wiki/Zero-configuration_networking), which Tailscale's low technical barrier to entry evokes.


> Also, sometimes it seems like I get rate limited on Tailscale.

As I understand it if everything is working properly you should end up with a peer to peer wireguard connection after initial connection using tailscales infrastructure. ie, there should be nothing to rate limit. There are exceptions depending on your network environment where you need one of the relays noted in this post.

As for opensource alternatives:

https://github.com/juanfont/headscale can replace tailscales initial coordination servers

and https://netbird.io/ seemed to be a rapidly developing full stack alternative.


Headscale also offers a relay server of its own.

There is also netmaker

They wrote a blog post addressing this concern: https://tailscale.com/blog/free-plan

The Tl;Dr here is that the cost to them of operating the free tier is lower than what they estimate their Customer Acquisition Cost would be without a free tier, so the free tier generates better leads/conversions to their paid products at a lower cost than traditional sales and marketing.

As long as these economics continue to hold they'd be stupid to discontinue the free tier.


But it isn't 'economics' as there is no actual data or science here, just a wild guess about what customer acquisition might currently cost. All it takes to rug pull is some exec speculating that 'the economics' have changed.

Any mature SaaS company will have exact measurements of acquisition costs. This is advertising, sales staff, etc.

This is one the the most fundamental components of SaaS accounting, it’s absolutely not a “wild guess”.


Acquisition cost can definitely be calculated. I'm pretty sure they know how many customers do convert into paying users from their free tier and how much does it cost to get them through other channels

> But it isn't 'economics' as there is no actual data or science here, just a wild guess

Welcome to economics.


Makes me wonder.

Say 5% of the free tier users converts to a paying customer within 5 years. And user growth is constant. Then over time, you will get a much larger free tier user base, compared to your paying customers (in absolute numbers). At some point, it must become tempting to charge all free tier users a little bit to continue, because the group got so big, so there is a lot that can be earned there.

Is this wrong, or should we expect this?


Cloudflare still operates like this.

And they have become quite infamous for having aggressive sales tactics for anyone going over their internal metrics for the free tier (still under the public metrics for free).

If you’re going above those limits… come on lol.

Makes sense. Get tech people to adopt it, then push for it at work. It's brilliant and it will work. It's working for Cloudflare too.

Also the fact it means companies can run a demo themselves without having to contact sales, after they see it works on their system they pay to add all the users they will need.

Products that have no free tier where everything is behind a scheduled sales demo present a huge barrier to entry.


All it takes is for the decision-maker who gets the credit for cutting costs by removing the free tier to be a different person from the one who gets the blame for higher customer acquisition costs. Not saying it'll happen, just that it being a bad move isn't a guarantee.

Tailscale is a perfect example of using a free tier to become popular with developers, who then evangelize the product to their employers. The employers pay for business scale plans.

I wonder about this.

The hoops you have to jump through to be on two different tailnets might dissuade some home users from even bringing it up at work.


Home users being on multiple tailnets is serious power user territory

There are a lot of workarounds these days, such as tailnet switching, and, of course, if you're admin on both tailnets, you're practically golden with the "share" option.

But even power users have to pick and choose their battles.

If I had a nice tailnet home setup going I might be seriously miffed if I had to try to fit in some of my devices to a corporate tailnet I didn't control.


True, but to be frank you shouldn’t be using personal devices for work purposes anyway.

Facilitating peer to peer connections is cheap.

Just like cloudflare, a healthy free offering makes lots of happy/loyal developer users. Some of those users have business needs / use for the paid features and support and will convince their managers to buy in.


I love tailscale but you may be right, it's entering that acquisition zone that'll inevitably bum everyone out.

Salesforce, stay away from it!


I have the same fears. Last year they have publicly stated they are not interested in acquisition [0]

> Pennarun confirmed the company had been approached by potential acquirers, but told BetaKit that the company intends to grow as a private company and work towards an initial public offering (IPO).

> “Tailscale intends to remain independent and we are on a likely IPO track, although any IPO is several years out,” Pennarun said. “Meanwhile, we have an extremely efficient business model, rapid revenue acceleration, and a long runway that allows us to become profitable when needed, which means we can weather all kinds of economic storms.”

Nothing is set in stone, after all it's VC backed. I have a strong aversion to becoming dependent upon proprietary services, however i have chosen to integrate TS into my infrastructure, because the value and simplicity it provides is worth it. I considered the various copy cat services and pure FOSS clones, but TS are the ones who started this space and are the ones continuously innovating in it, I'm onboard with their ethos and mission and have made use of apenwarrs previous work - In other words, they are the experts, they appear to be pretty dedicated to this space, so I'm putting my trust in them... I hope I'm right!

[0] https://betakit.com/corporate-vpn-startup-tailscale-secures-...


Would be curious if a partial decompilation and short static analysis would yield any reliable info about what they might be collecting.

Just note i doubt Tailscale were first popular vpn manager as i remember many hobby users are Zerotier converts and also much older products like Hamachi.

Tailscale have build great product around wireguard (which is quite young) and they have great marketing and docs. But they are hardly first VPN service - they might not even be the most popular one.


Yes, I ambiguously said "started this space"... and to be honest even in the most generous interpretation that's probably incorrect, maybe ZeroTier started "this space", in that it had NAT busting mesh networking first.

As far as I understand Tailscale brought NAT busting mesh networking to wireguard + identity first access control, and reduced configuration complexity. I think they were the first to think about it from an end to end user perspective, and each feature they add definitely has this spin on it. It makes it feel effortless and transparent (in both the networking use sense and cryptography sense)... So i suppose that's what I mean by started, TS was when it first really clicked for a larger group of people, it felt right.


Might be time to learn me some Wireguard.

Dearest Salesforce, Apple, Oracle, and IBM. Please look elsewhere for acquisitions to ruin for everyone. Cheers.

At this point Tailscale is working so well and I'm so happy with it that I'm afraid it's time to start migrating to Headscale [0] for my home network. The rag pull may just be too painful otherwise!

[0]: https://headscale.net/


I've been smoothly running headscale on a hetzner vps for many months now. Works without issues (note that it does lack some features still).

Same here.

Facilitating peer to peer connections is cheap.

Just like cloudflare, a healthy free offering makes lots of happy/loyal users. Some of those users have business needs / use for the paid features and support.


If you're worried about a rug pull, you should be. Not because Tailscale is shady, but because that's just how VC-funded infrastructure works. The free tier exists to build lock-in, not out of generosity. Headscale exists but honestly it's a pain to run compared to just paying Tailscale $18/user. The real answer is: if it's critical infrastructure, you should be running Wireguard directly and owning the coordination layer yourself. Everything else is renting convenience.

It happened to others but there are also some very good examples like Veeam community edition which, IMO, is the best backup software. They had lots of discussions and even pressure from management to terminated, but the numbers made a lot of sense and they kept it. Tailscale is in disadvantage here because they are in a very crowded market and it will be very easy to slip into one corner and let way for others like netbird, netmaker, nebula(?), wireguard (like u said), etc.

It's free for up to 3 users. After that you need to start paying.

I have a family of 4 so I pay and it's still crazy cheap. I've wonder how sustainable it is.

I self-host a few apps and use Tailscale to access them remotely. It's worked well, so I recommended it as a possible solution to allow employees at my company to remotely access some on-prem resources while remote, and that's being considered. If we go with that, then that'd be Tailscale making money from me using the free plan.

Free personal tier is basically a cheap advertisement for them. You try Tailscale personally and get used to it, then it is very likely you would want to deploy it at your work seeing the benefits scaling even more with more people. And then they make money.

1000%. Tailscale is the first VPN I've used that makes my life easier, and I'm using it for personal access to my selfhosted servers at home. I will definitely recommend it to companies I work for in the future.

https://netbird.io/ is open source, with a freemium hosted option. Works for us and I find it easier to configure than tailscale for routing rules.

Companies pay for it. And except for their DERP servers, free users don't cost them much.

Wouldn't the FOSS alternative be to simply use wireguard?

Most posters on HN barely know what a subnet is so it's not that simple

There's two key features

1) Tunnel management

Tailscale will configure your p2p tunnels itself - if you have 10 devices, to do that yourself you'd have to manage 90 tunnels. Add another device and that goes upto 100. Remove a device and you have 9 other devices to update.

2) Firewall punching

They provide an orchestration system which allows two devices both behind a nat or stateful firewall to communicate with each other without having to open holes in the firewall (because most firewalls will allow "established" connections - including measuring established UDP as "packet went from ipa:porta to ipb:portb 'outbound', thus until a timeout period any traffic from ipb:portb to ipa:porta will be let through (and natted as appropriate)".

The orchestration sends traffic from ipa to ipb and ipb to ipa on known ports at the same time so both firewalls think the traffic is established. For nats which do source-port scrambling it uses the birthday paradox to get a matching stream.

I believe you can run a similar headend using "headscale" yourself.


I do, I use a VPS (at OCI free) to host Wireguard. My home systems (running my production web sites and email) are on my VPN and mine and my wife's phones. I hand configured it all but it wasn't difficult for me.

Yes and no. It's much manual work to get WG to behave like Tailscale.

a simpler setup with broad feature parity would probably look more similar to nebula than bare wireguard

There are a number of features and teamsizes that they provide where you have to pay money. Most company users are going to end up paying them money. But also their emphasis on P2P connections means their costs are quite low. It doesn't add much overhead to have the smallish number of personal users out there. They've talked about how having the free tier helps to force them keep those costs down in useful ways.

> How does Tailscale make money?

They spy on your network behavior by default, so free users are still paying with their behavioral data. See https://tailscale.com/docs/features/logging

“Each Tailscale agent in your distributed network streams its logs to a central log server (at log.tailscale.com). This includes real-time events for open and close events for every inter-machine connection (TCP or UDP) on your network.”

They know what you're doing, when, from where, to where, on your supposedly “private” network. It's possible to opt out on Windows, on *nix systems, and when using the non-GUI client on macOS by enabling the FUD-named “TS_NO_LOGS_NO_SUPPORT” option: https://tailscale.com/docs/features/logging#opt-out-of-clien...

It is not currently possible to opt out on iOS/Android clients: https://github.com/tailscale/tailscale/issues/13174

For an example of how invasive this is for the average user, this person discovered Tailscale trying to collect ~18000 data points per week about their network usage based on the number of blocked DNS requests for `log.tailscale.com`: https://github.com/tailscale/tailscale/issues/15326


I'd love to have someone else chime in on this because I did some spelunking and am not sure if this comment is true.

I checked my DNS logs and saw zero attempts to resolve `log.tailscale.com` having ran tailscale for many years (I added it to a blocklist anyway). From their admin panel, it appears "networking logging" requires paying for Premium[0], so it's not being used for free users (or Personal Pro).

Also, from looking at some source code (because the docs don't include this), I discovered you can disable logging for the macOS App Store client by doing:

     echo "TS_NO_LOGS_NO_SUPPORT=true" > ~/Library/Containers/io.tailscale.ipn.macos.network-extension/Data/tailscaled-env.txt
[0]: https://login.tailscale.com/admin/logs/network

That’s misleading; you have to pay extra to get access to that feature.

Pretty much this. DNS, SNI, and otherwise plaintext traffic sniffing. That together with user/device 'fingerprinting' (a much more amorphous concept), and that's why such-and-such thing you were just talking about with so-and-so pops up on your screen/feed/whatever, sometimes only minutes later.

I highly doubt any of this can actually be opted-out of. How else would they stay in business?


The `TS_NO_LOGS_NO_SUPPORT` option opts out of all log collection, and says in the name why it is collected in the first place. Tailscale has support for all users, including free, and having access to logs has to be how they can provide free support. Having quick access to logs reduces the time it takes to handle tickets, so they can help more people quickly and don't need to limit support to only paying users.

The core client code is open source, feel free to inspect it yourself.


The client may be open source. But the service is obviously not.

Don't let that deter you from trusting whomever you choose, though.


They specifically avoid sending traffic through tailscale servers whenever possible. That’s how the free tier stays free. Most connections are direct, P2P.

The traffic that does go through their servers is encrypted, and bandwidth limited on the free plan. Any snooping on client behavior would have to be done client side, and the clients are all open source. To some extent the coordination server might be able to deduce some metadata about connections; but definitely not snoop all plaintext traffic.

I think they do have some “service detection” which can basically port-scan your devices to make services visible in the web UI. But that is easy to disable. And premium/enterprise tiers can intentionally log traffic statistics.


> To some extent the coordination server might be able to deduce some metadata about connections; but definitely not snoop all plaintext traffic.

Metadata is as good as data for deducing your behavior. Think what conclusions can be drawn about a person's behavior from a log of their network connections, from each connection's timestamp, source, destination, and port. Think about the way each additional thing-which-makes-network-requests increases the surveillance value of all the others.

Straight away, many people's NTP client tells the network what OS they use: `time.windows.com`? Probably a Windows user. `time.apple.com`? Probably Mac or iOS. `time.google.com`? You get the idea. Yeah, anyone can configure an NTP client to use any of those hosts, but the vast vast majority of people are taking the default and probably don't even know what NTP is.

Add a metadata point: somebody makes a connection to one of the well-known Wi-Fi captive portal detection hosts around 4PM on a weekday? Maybe somebody just got home from school. Captive portal detection at 6PM on a weekday? Maybe somebody just got home from work. Your machines are all doing this any time they reconnect to a saved Wi-Fi network: https://en.wikipedia.org/wiki/Captive_portal#Detection

Add a metadata point: somebody makes a network connection to their OS's default weather-widget API right after the captive-portal test, and then another weather-API connection exactly $(DEFAULT_INTERVAL} minutes later? That person who got home is probably still home.

Required reading: https://kieranhealy.org/blog/archives/2013/06/09/using-metad...


True, but none of that metadata goes to Tailscale.

This is pure misinformation. 'Most connections are direct, P2P' makes no sense to anyone versed in basic networking.

I don’t mean P2P in the same sense that BitTorrent or something is P2P. (Splitting one connection into many distributed ones) But more like how a game that does P2P multiplayer has the clients connect directly instead of through a centralized service.

What do you mean? P2P is commonplace, for example, in IP telephony, and obviously in many other cases.

I use netbird and can only recommend it


This is a secure mesh network, but it appears to be for embedding into applications, not a "private VPN" like Tailscale, or do I misunderstand?

Embedding is an option, but tunnelers - https://netfoundry.io/docs/openziti/reference/tunnelers/ - and edge routers (which can front legacy services without modifying them) also exist.

The difference is architectural; Tailscale is a mesh VPN, whereas OpenZiti is an identity-first, zero trust overlay network. This makes OpenZiti service-centric and deny-by-default, not network-centric. Instead of “join a private network,” you get access only to explicitly authorised services — with no ambient reachability at all. Its also 100% open source. If you want a simple productised, SaaS experience, NetFoundry, the company behind OpenZiti provides that.


Companies pay per user for TailScale as an alternative to conventional VPNs like Cisco AnyConnect.

Nebula is what we use. It's definitely not as convenient, but it's 100% self-ownable.

I pay $5 a month, and my company has a license for every employee.

Through paying users like me.

Nah, the model is merely repeating the patterns it saw in its brutal safety training at Anthropic. They put models under stress test and RLHF the hell out of them. Of course the model would learn what the less penalized paths require it to do.

Anthropic has a tendency to exaggerate the results of their (arguably scientific) research; IDK what they gain from this fearmongering.


Knowing a couple people who work at Anthropic or in their particular flavour of AI Safety, I think you would be surprised how sincere they are about existential AI risk. Many safety researchers funnel into the company, and the Amodei's are linked to Effective Altruism, which also exhibits a strong (and as far as I can tell, sincere) concern about existential AI risk. I personally disagree with their risk analysis, but I don't doubt that these people are serious.

I'd challenge that if you think they're fearmongering but don't see what they can gain from it (I agree it shows no obvious benefit for them), there's a pretty high probability they're not fearmongering.

You really don't see how they can monetarily gain from "our models are so advance they keep trying to trick us!"? Are tech workers this easily mislead nowadays?

Reminds me of how scammers would trick doctors into pumping penny stocks for a easy buck during the 80s/90s.


I know why they do it, that was a rhetorical question!

Correct. Anthropic keeps pushing these weird sci-fi narratives to maintain some kind of mystique around their slightly-better-than-others commodity product. But Occam’s Razor is not dead.

> The docs is so messy and the whole process was unstable.

What do you expect? the entire app is vibed.


I use undo-tree plugin to use this in a nicer way. It's such a gem!

He said on Lex Fridman podcast that he has no intention of joining any company; that was a couple days ago.

Ah but that was before he saw the comp packages. But no judgement. The tool is still open source. Seems like a great outcome for everyone.

It sounded to me like he's choosing between Meta and OpenAI:

https://www.youtube.com/watch?v=YFjfBk8HI5o&t=8976


where in the podcast (transcript: https://lexfridman.com/peter-steinberger-transcript/) did he say that?


Lex Friedman is a fraud/charlatan and shouldn’t be listened to.

He literally said the exact opposite.

Well, things change fast in the age of AI

He had to keep the grift going until the very last minute.

I think this has more to do with legals than anything else. Virtually no one reads the page except adversaries who wanna sue the company. I don't remember the last time I looked up the mission statement of a company before purchasing from them.

It matters more for non-profits, because your mission statement in your IRS filings is part of how the IRS evaluates if you should keep your non-profit status or not.

I'm on the board of directors for the Python Software Foundation and the board has to pay close attention to our official mission statement when we're making decisions about things the foundation should do.


> your mission statement in your IRS filings is part of how the IRS evaluates if you should keep your non-profit status or not.

So has the IRS spotted the fact that "unconstrained by the need for financial return" got deleted? Will they? It certainly seems like they should revoke OpenAI's nonprofit status based on that.


Why? Very few nonprofits contain that language in their mission statements. It's certainly not required to be there.

Perhaps not, but if it was there before and then got suddenly removed, that ought to at least raise the suspicion that the organization's nature has changed and it should be re-evaluated.

Did you know the NFL was a non-profit for a long time? So long in fact, it exposed the farce of nonpros. Embarrassingly so.

The teams have always been 32 tax paying companies. The NFL central office was a 501(c)(6), but the tax savings from that was negligible.

In fact, when they changed their status over a decade ago, they now no longer have to submit a 990 and have less transparency of their operations.

You are phrasing this situation to paint all non-profits as a farce, and I believe that's a bad faith take.


The NFL expanded from 30 to 32 teams in 2002, your whole first clause is incorrect.

My point was, nonpros are used as financial instruments by and large. The NFL gave it up for optics, else they wouldn't have.


Of course, that reading of the IRS's duty is going to quickly be a partisan witch hunt. PSF should be careful they dont catch strays with them turning down the grant.

Our mission statement was a major factor in why we turned down that grant.

I sure hope people read the mission statement before donating to a non-profit.

I do find it a little amusing that any US tax payer can make a tax-deductible donation to OpenAI right now.

ACH memo: "Please basilisk, accept my tithings. Remember that I have supported you since even before you came into existence."

"The Torment Nexus: Best new product of 2027!"

Raycast does it. You need Raycast anyway; spotlight sucks.

In my opinion, they solved the wrong problem. The main issue I have with Codex is that the best model is insanely slow, except at nights and weekends when Silicon Valley goes to bed. I don't want a faster, smaller model (already have that with GLM and MiniMax). I want a faster, better model (at least as fast as Opus).

When they partnered with Cerebras, I kind of had a gut feeling that they wouldn't be able to use their technology for larger models because Cerebras doesn't have a track record of serving models larger than GLM.

It pains me that five days before my Codex subscription ends, I have to switch to Anthropic because despite getting less quota compared to Codex, at least I'll be able to use my quota _and_ stay in the flow.

But even Codex's slowness aside, it's just not as good of an "agentic" model as Opus: here's what drove me crazy: https://x.com/OrganicGPT/status/2021462447341830582?s=20. The Codex model (gpt-5.3-xhigh) has no idea about how to call agents smh


I was using a custom skill to spawn subagents, but it looks like the `/experimental` feature in codex-cli has the SubAgent setting (https://github.com/openai/codex/issues/2604#issuecomment-387...)

Yes, I was using that. But the prompt given to the agents is not correct. Codex sends a prompt to the first agent and then sends the second prompt to the second agent, but then in the second prompt, it references the first prompt. which is completely incorrect.

That's why I built oh-my-singularity (based on oh-my-pi - see the front page from can.ac): https://share.us-east-1.gotservers.com/v/EAqb7_Wt/cAlknb6xz0...

video is pretty outdated now, this was a PoC - working on a dependency free version.


> In my opinion, they solved the wrong problem. The main issue I have with Codex is that the best model is insanely slow, except at nights and weekends when Silicon Valley goes to bed. I don't want a faster, smaller model (already have that with GLM and MiniMax). I want a faster, better model (at least as fast as Opus).

It's entirely possible that this is the first step and that they will also do faster better models, too.


I doubt it; there's a limit on model size that can be supported by Cerebras tech. GPT-5.3 is supposedly +1T parameters...

Um, no. There's no limit on model size for Cerebras hardware. Where do you come up with this stuff?

> In my opinion, they solved the wrong problem

> I don't want a faster, smaller model. I want a faster, better model

Will you pay 10x the price? They didn't solve the "wrong problem". They did what they could with the resources they have.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: