Jacob Hess

I love tech, product strategy & platform innovation. I focus on engineering, automation & cutting edge tech. My free time consists of ultras, the fam & stogies!

[Jacob Hess @Jacobhess]

@JoshuaBarzon I’ve always called it the St Louis Arch or Gateway Arch, or, just the Arch.

If I’m not mistaken they changed the name roughly a decade ago.

Fun fact, it’s also the smallest National Park!

[Jacob Hess @Jacobhess]

For Daybreak access, there are extra security measures on your account but honestly they aren’t too overbearing.

The only pain is sometimes a remote headless instance inside will expire, but, it’s pretty minimal effort to repair.

Totally worth signing up.

Side note, there’s a deal with OpenAI and Yubikey for cheap hardware 2fa if you chose to go that route.

[Tibo @thsottiaux]

Our team worked with OpenCode to have their users who are also OpenAI API customers back in business and any credits refunded which would have been drawn in the attack.

Also, you can request Daybreak Blue access to use our cyber permissive setup for defense. Do not wait until you have to deal with an incident. We’re here to help.

[Jacob Hess @Jacobhess]

Uh, wow.

Elon’s “The odds we’re in base reality is one in billions.” Maybe one in trillions now?

[Odyssey @odysseyml]

We placed the fruit fly brain inside Odyssey-3, a frontier world model.

The lil’ guy is now flying around totally imagined, simulated worlds all by itself 🫶

[Jacob Hess @Jacobhess]

I haven’t used MJ since nano banana’s release…I’m going to have to dust it off and give it a whirl!

Forgive me (as you probably have explained it on numerous occasions, I’ve had to pause on listening to a lot of top tier pods due to being stretched a bit thin) but why mid journey instead of the other image models?

[Jacob Hess @Jacobhess]

Take this with a grain of salt, this was a quick chat w/ Grok.

—

TLDR

Using Jev’s image analysis solely to identify or classify whether content is NSFW (adult) does not appear to violate the published Acceptable Use Policy based on a plain reading of the text.

Extra Context

The governing document is TypeSafe’s **Acceptable Use Policy** (last updated September 23, 2026), which is incorporated by reference into the Master Customer Agreement. Relevant portions:

### Child safety (explicit prohibition)

> **2.1** create, share, or promote sexually explicit material involving minors, content that exploits or sexualizes children, or age-inappropriate content targeting minors (e.g., child sexual abuse material (CSAM), sexual material, graphic violence, or obscenity);

This is limited to minors. It does not address adult content.

### Obscene / sexual exploitation clauses

> **1.7** engage in, promote, support, or facilitate obscene, defamatory, hateful, discriminatory, threatening, or violent activities, whether online or offline, including to disclose sensitive information without permission or incite violence to people or animals;

> **1.12** facilitate human trafficking, prostitution, sexual violence, or other forms of exploitation;

These focus on *engaging in, promoting, supporting, or facilitating* the listed activities (or creating/sharing related content). Classifying or scoring an uploaded image to determine whether it is adult/NSFW for moderation purposes is not described as engaging in or facilitating those activities.

### Other relevant framing

The AUP states it is “not exhaustive” and that TypeSafe “reserves the right to take remedial action in connection with content or uses not specifically described below.” It also says customers must not use the Services to “upload, process, store, or attempt to create any illegal content” (1.2) and that customers are responsible for their Input.

The Master Customer Agreement further requires compliance with the AUP and makes the customer responsible for Input content and legal compliance.

### Bottom line

- There is **no explicit ban** on analyzing or classifying adult sexual content.
- The prohibitions target creation, promotion, facilitation of illegal/exploitative sexual content, and anything involving minors.
- Pure identification/moderation of adult NSFW content does not clearly fall under the listed prohibitions.

Because the AUP is not exhaustive and uses terms like “obscene” and “facilitate,” TypeSafe could still interpret a high-volume moderation use differently or update the policy. For production use, the clearest confirmation would come directly from TypeSafe. Image support itself is also still marked experimental (“danger zone”).

[Jacob Hess @Jacobhess]

Am I missing something in how I’m using dot, or are these early integration issues?
I’ve started giving dot real tasks that I normally handle through ChatGPT/Codex. I’m optimistic about the idea and want to share useful feedback, because several familiar workflows have required much more supervision than I expected.
A few examples from roughly the last day:

1. Apple Messages discovery and execution
I thought there was an official Apple Messages plugin, so, I suggested this to Dot. Dot said there wasn't an official path, so it focused on computer use and third-party options. 8 hours later, dot confirmed that the official plugin was bundled with my app but wasn’t installed (i guess just proactive research, i didn't reprompt it to check again).
For that message, I was asking it to schedule dinner questions to my kids for 6:30 a.m. During the computer-use attempt, it scheduled one for the wrong time, then accidentally sent it late at night while editing. It then undid the send...

2. Follow-through on the original request
Dot told me neither message was successfully scheduled. I then asked it to pause computer use overnight (because i didn't want more whoopsies) and pick things back up during the 5 a.m. hour. I expected the morning request to remain part of that work.
It resumed, encountered a locked Mac, and reported that blocker. I have the settings enable to use my mac while locked.
Anyway, the 6:30 a.m. messages were never sent, and the original time-sensitive goal got lost (according to the dot due to troubleshooting).

The feedback here is about keeping the original outcome and deadline visible, and clearly escalating “this won’t happen” while I still have time to do it myself.

3. A checklist that didn’t work on my phone

Dot created an ingredient checklist in a Page so I could check my pantry and fridge. I couldn’t operate the checkboxes on mobile and ended up pasting my inventory back into chat. The checklist existed, but the interaction I needed wasn’t usable on the device where I needed it.

4. Browser access and confusing task state

I’ve successfully ordered groceries through Codex several times. This attempt became an extended browser troubleshooting session.

I explicitly asked for the task’s internal browser so it wouldn’t interfere with my work. Instead, I repeatedly saw a local browser (Arc) access prompt, which I denied several times. The task also appeared hung enough that I couldn’t type into it.

The threads looked cloud-based to me, even after I requested my Mac and a specific local project. Investigation eventually verified that the task was running on my Mac.

One attempt added two grocery items, then stopped after ERR_NETWORK_CHANGED. A fresh local-project thread initially couldn’t find its internal browser either.
Opening the task through the app restored access. The cause of the initial connection delay is still unresolved, and the order wasn’t complete when I wrote this.

What I’d love to see improve:
- Reliable discovery of the official tools already available
- Clear execution location and browser status
- Stop instructions and permission denials that take effect predictably
- Recovery from routine browser failures without me repeatedly supervising
- Time-sensitive commitments that stay attached to their original deadlines
- Parent-task visibility that distinguishes active work from a stalled task

I’m not angry; I know these are early days. I’m sharing this because the concept is valuable, and these workflows have already worked well for me in regular Codex.

I’d love dot to carry that reliability forward while reducing how much I need to manage.

[Jacob Hess @Jacobhess]

Am I doing something wrong w/ my Dot?

I finally decided to give it a real try, and, to have it do things instead of my normal chatgpt/codex flows.

Here's a few problems, just in the last 12ish hours:
- I had suggested that OpenAI created a plugin for Apple Messages. It said nope, we had to use computer use. So, I suggested to just schedule a message for 630a. It proceeded, then, messed up computer use and sent my 9 year old a text (to his apple watch) at 11p, then, deleted the message a short while later since it was a mistake. I told it no worries, just send the message at 630a manually (using computer use).
- I checked at 730a this morning, and, the dot just didn't send the message... When asked it said opps and appolgized.
-This morning it also said, "opps, i guess open AI does have a plugin, my bad".

Later this morning, I asked it to order groceries (something I do frequently in codex), and, it's been more troubleshooting:
- it created a checklist (for my grocery list) in spaces so i could see what I had in the fridge and pantry, but, checklists don't work on mobile for pages in spaces.
- it keeps trying to start local threads, but, they are cloud threads.
- the first cloud thread couldn't use the browser. After assuring it that it has a browser, it finally worked. Then, two items in it received "ERR_NETWORK_CHANGED" and gave up. I'm not sure why the network would've changed, but, it just quit.
- the second thread also didn't have access to a browser, until I reminded it that it did, then, it leveraged it.

I want to be optimistic, i think the idea is really good, but, it's taking steps backwards that we've already had fixed in Codex/ChatGPT for months now.

[Jacob Hess @Jacobhess]

I agree, but, at some point, as a consumer, we feel as if we’re overpaying for what we get, and I think due to the popularity of these type of rentals a lot of investors overpaid to get in the door, then have to make it up with things like cleaning fees.

My personal problem wasn’t with the owners per se, but the property managers.

About a year ago, I had two issues back to back with Airbnb (on one vacation). The first one ended up repairing the problem in about a day, but the other one ended up submitting a $1,500 claim against us for a $750 telescope (that we couldnt ever get to work when we showed up).

At that property, the hot tub that was advertised. Broken. Like, smelly dirty half full water broken. Not a few days broken. No notice, not even attempted to be fixed in the four days we were there.

The pickleball court? So many large cracks it was unplayable. Not pictured.

The stairwell from the entrance to the main floor, the lights were out and they literally had a plug in cheap star led projector that you had to use for any light.

The dishwasher didn’t work.

One of two showers didn’t work.

The starlink internet didn’t work (I got it to work after about 30 minutes of troubleshooting with tech support, ultimately moving the satellite receiver). This was only critical because there wasn’t good cell service where it was at.

I could go on and on, but, bottom line this wasn’t a cheap Airbnb. This was a premium property that we booked as a special treat for our kids, and a lot was wrong.

We couldn’t figure out how it was 5 star reviews and no mention of the multitude of problems. We left a four star review, because, despite all the problems we still felt like it had a lot of positives, we just called out our experience.

Then, a few days later the $1,500 charge was sent to us.

After spending about a week going back and forth with AirBNB, they scolded us and told us that we would have a mark against our record, even though the property couldn’t manage to even send a picture of the part that “we broke”.

We’ve had probably 25 or 30 stays over the years, with nothing but 5 star reviews. We’re honestly pretty easy going, but you need dishwashers and laundry machines to work when you’re at a place for an extended period of time with three kids.

Guess what. About 2 months later we got an email from the property manager. If we would remove our review they would send us a $500 gift card.

We mentioned it to Airbnb, they never responded.

That was my last straw. I went back to hotels.

I’m not saying I’d never use an Airbnb, but, it’s definitely more of a last resort than a first choice.

So, it’s kind of a two way street. I think you are paying the price for scammy Airbnb hosts, just as we are paying for scumbag renters. At some point, we both give up.

A hotel has the manpower to deal with it. I don’t think most Airbnb owners do, and then they end up losing their hard earned dollars on the investment.

[Jacob Hess @Jacobhess]

For a large majority of deals, probably not needed.

But I’ve found that people get crazy about a couple thousand of dollars somewhere along the way, and get on some moral high ground about how the other party is negotiating/acting, and then they blow up a perfectly sound deal.

One house we sold, we accepted an offer for $15k less than the top offer because it was our neighbors who wanted their dad to live next door and we liked them from being neighbors over the years.

They chose not to use an agent, so, we cut our agents commission to be only our portion of the deal.

All in all we still were losing money in comparison to the top offer, but, at 3% of reduced commission, it was pretty close to even.

Then, after inspections they tried to get us to come down another $15k because “they already were overpaying, so the house had to be perfect.” (They had bought their house for about 30k less two years before, but this was when homes values were going through the roof.)

They weren’t overpaying, we knew because we also were buying and knew the market pretty well.

We tried to be kind, and offered to fix roughly half of the items that were reasonable to fix, but not the purely cosmetic stuff. Even then, it felt like $7.5k was too much, like, that they could get everything, including the cosmetic items fixed for about that, but we had another house under contract so we were just trying to make it work, and didn’t have time to hire it all to be done.

And it’s not that we were letting things go at the house, we had put about 15% of the homes value into improvements over the 4 years we lived there.

Anywho, they made a big deal about it, stuck to their guns ($15k or nothing) and we ended up agreeing to cancel the deal.

Our agent reached out to all the original offers, and within 24 hours we secured a deal with an agent and their client for roughly the largest originally offered amount, disclosed the items on the inspection report, and they bought it with no fixes.

Bottom line, the house was priced fairly, (maybe even a little low because we had another house we were trying to buy and were trying to move quickly).

Did the other agent earn their keep? I’m not sure, but, they got the deal closed. They knew the market, knew what was fair, and worked through it with their client.

One family could have bought our home for nearly $25k less, but, couldn’t get out of their own way. Maybe in the long run it was for the best for them, but, if they wanted the house, they failed at their job.

Take this with a grain of salt, but now, 6 years later, Zillow says that house is worth about $150k more than we sold it for.

[Jacob Hess @Jacobhess]

I resonate nearly everything you wrote, but most of all with your final paragraph. I’ve come to realize loyalty doesn’t mean anything at my level of subscription or social media standing.

I was lucky enough to be in on Dall-e, and, signed up for chatGPT the day I could. I paid money the day I could. I’ve been Pro-200 since the day I could.

Was I disappointed I didn’t get invited to either model release party? Sure. I was even planning on paying my own airfare and hotel.

Was I disappointed when social media folks get models weeks ahead of time (or historically when model roll out wasn’t as efficient, and we’d even get them days after release)? Sure, but that’s because I love the company and product so much, i just want to be all in.

I literally was ready to go to pro-500 the day I could, then, I hesitated. 20x v 25x for 2.5x more…that doesn’t feel quite right, especially because pre Astra, i felt like I couldnt hit my monthly cap, and, now I can’t last a few days on it, but you want me to shell out $500 for 3 days of usage?

I know it’s subsidized, and I’m getting more value than I pay for (at least considering api cost), but the straw that broke the camels back was when I used Astra + sub agents with less intelligence and blew through 2 resets on a pretty straightforward project, with poor results. I tried the same project on Opus 5.5, and it did it well on only ~20% of the $100 plan I have through work.

I love hype, it lets me know what to aim for, but right now either I’m doing something wrong or it’s being overhyped.

Sometimes I think the really killer hype results probably come from the insiders having an unlimited token budget. I’m sure if I could run Astra at an unlimited pace I’d be really impressed, but I’ve had to quit using Astra if I want my $200 plan to last more than a couple of days.

All that being said, I’m still an open ai fan, I have no plans to cancel my 200 plan, but unfortunately for my situation I think I’m not really their target. Maybe if I bought 5 or 10 200 plans, or had a social media following I would be their market, I’d probably get more perks, but I don’t think loyalty really matters to them.

I believe they still do great work, and if I look at what we have access to, and rewound 5 years Imy mind would be melted. I think openAI will get back on top soon, but your post resonated with me because I’m at a low on the current state of things with them right now.

[Jacob Hess @Jacobhess]

@Jason @bot @XMoney 🤑

[Jacob Hess @Jacobhess]

@daniel_mac8 Ahh. I guess I didn’t ever notice that it didn’t adapt. I probably just type and switch threads and never noticed!

[Jacob Hess @Jacobhess]

Hey @PatrickMahomes, I gotta talk to you about your Whataburgers. We are now 2 for 2.

Two different locations, two different times.

Two great experiences.

Excellent customer service, excellent food, excellent ketchup.

Brae specifically was the gal that helped us today. Please pass along our thanks, we’re planning to stop again next time we’re in KC.

[Jacob Hess @Jacobhess]

@poteto @technoSaas It took until this comment for me to understand what you meant, haha.

“Bro, have you tried the latest O-P-U’s…? I think it’s 5.5?! It’s such a whiny try hard, but at least it doesn’t take a stream of resets to use!”

[Jacob Hess @Jacobhess]

Once I paid whatever they demanded for “unlimited” because I was testing prompts for small business images with consistent models for an ad.

Come to find out, they made it so that you could only produce images at such a rate that you literally would’ve veen cheaper to just pay nano banana api cost ($0.13/image, if I remember correctly).

Fool me once, shame on you. Fool me twice, shame on me.

[Jacob Hess @Jacobhess]

The cookies are huge and taste great but at $4.50 (before tip) it is a lot for a single cookie.

Yes, I know you can get mini’s in a 3 pack for $9.99 but it’s basically the same “volume” of cookie as a single, you just pay 2x.

Oddly, last year I made a lot of chocolate chip cookies for a competition and I could get crumbl clone chocolate chip cookies made for about $1.85, (with high quality chocolate, which I only mention because the chocolate is like 2/3 of the total cost).

I would think most of their cookies don’t cost nearly that to make (chocolate and vanilla happen to be an expensive), and they probably get a much better deal on chocolate than I do going and paying full price at the grocery store for a premium brand.

Long story short, (as I truly wish them the best, and I understand they need to pay for the store, employees, franchise fees, etc, but) it’s just tough to justify spending premium on what the kids see as “a cookie” for my family of 5.

I’m probably not the target demographic though, haha.

[Jacob Hess @Jacobhess]

@OpenAI For reference, this is for an empty test list in a todo app that we're trying to delete.

I understand that we don't want to accidentally delete things that are tough to recover, but, the approval service has been a bit of a thorn in my dot usage so far.

[Jacob Hess @Jacobhess]

It seems like @OpenAI Dots have an approval service that runs before interacting w/ webpages. It's been finicky in two different situations in the last day, but, this one was a little unexpected:

"The move is paused: this time the approval service explicitly says it’s at capacity and didn’t execute the action. [Folder] is unchanged; I’ll leave the move queued."

[Jacob Hess @Jacobhess]

I’m really pro codex and models. I guess o1 and o3 got me fully on OpenAI and it’s tough to switch my habits. That being said, I’ve had the same struggle.

- I had Astra and Sol 5.6 flub a sizeable but straightforward project.
- Opus 5.5 did it, and, did it well.
- Codex found some errors in how the code would work with the rest of the environment (that Claude missed), Claude fixed them.

At least Sol 6.1 fixed the decimation of tokens / subscription drain, but is pretty slow, and my fear is artificially cheap.

For targeted tasks (even at some size) it seems to do well, but, i do feel like I haven’t quite figured out how to loop/ prompt it quite right, while opus seems to just get it.

Bottom line, I’ve felt the need to up my instructions / prompting because it’s getting expensive and time consuming to be wrong.