Tech Executive in Tokyo | 17+ yrs scaling products & teams across US/Japan | Ex-Founder (Techstars Tokyo) | Gamer đž & Heavy Metal Enthusiast đ¤
Tokyo, Japan
Joined December 2009
- Tweets237
- Following3.5K
- Followers775
- Likes841
Will Bohn retweeted
We need to talk about Anthropic. Its relationship with its power users appears to be collapsing in real time.
A few weeks ago, my post about Claude becoming lazy and failing at basic agentic tasks received 2.5 million views and more than 2,000 replies, with the overwhelming majority of commenters agreeing.
This weekend, Anthropic framed a 17% reduction in current weekly usage as a âpermanent 25% increase.â It deleted the original announcement, reposted it with an explanation, and still received a Community Note.
Today, users discovered that the $200 âMax 20xâ plan only provides 20x usage within a five-hour window. Its weekly allowance appears to be only roughly twice that of the $100 plan. Anthropic is already facing a proposed class-action lawsuit over these limits.
Meanwhile, Claude is going viral for ending a session after a user repeatedly insulted it.
Claude may be the only $ 200-a-month product where the agent regularly fails to finish the job, nobody can tell you exactly how much usage you have bought, and the product can decide that it no longer wants to work with you.
Employees like @trq212 take these concerns seriously and genuinely ask for honest feedback, which I really appreciate. I have always found Anthropicâs staff to be friendly and respectful.
But Anthropicâs public communication is very poor right now, and its models are also performing poorly in terms of practical usability. Fable is the exception, but its rate limits make it almost impossible to use consistently.
I am genuinely trying to understand the underlying issue, but I honestly cannot come up with a convincing explanation.
Anthropic set out to be the highly ethical AI company, a position that ultimately led to its clash with the DoD. But the idea that this philosophy should extend to Claude ending sessions because it objects to a userâs tone is baffling to me. Frankly, it is a deal-breaker.
An AI needs to be able to handle my frustration. After all, it is not a person who can actually be offended.
Anthropic is reportedly preparing for an IPO. Given that, it is even more puzzling that the company does not appear to be making an all-out effort to improve its public image.
Anthropic is making things remarkably easy for the competition right now.
All I can think of is Shodan from the System Shock game series. And yes, there are a lot of people out there like this...
Will Bohn retweeted
don't switch models to save money. set `delegation.model` to the cheap one instead.
your session keeps its cache, your subagents run cheap, and subagents are where most of the tokens go anyway :)
Just FYI. If you are switching models in a session all the time - you are doing it wrong.
Every time you switch, your entire prompt cache is invalidated on the new model you switch to, and you have to repay the full input tokens price for all of it.
Stop doing this unless those models are free.
This is not a hermes thing - this is a fundamentals of inference thing.
Will Bohn retweeted
Are you 40? If you are, it's time to finnally buy a WinRAR license! Thanks to @buraktuyan for creating this awesome video that no one asked for but that everyone needs! đ¤Łđ
Here's my 30-second spec ad for WinRAR (that nobody asked for).
Okay, perhaps Rourke Heath did.
He shared this fun brief with his GenHQ community last week, and I took it on.
In this experimental project, I wanted to turn one of the internet's longest-running jokes into a subtle family drama about WinRAR's 40-day trial that somehow lasted almost three decades.
Here's how I made it:
After writing the script (the old school way), I worked with ChatGPT 5.6 Sol Max to turn it into a 14,000-character prompt for Seedance 2.5 (yes, Dreamina allows prompts that long).
I used GPT Image 2 to create the reference image for WinRAR's trial screen, while Seed Audio 1.0 generated the audio reference for the full voiceover and dialogue. After 20+ Seedance 2.5 tests at 480p, with the prompt evolving throughout, I generated the final versions at 720p.
I then lightly edited the video and upscaled it to 1080p with Topaz Labs Astra Starlight Precision 2.6. However, the screenshot text didn't survive the upscale (I should have used Astra's scene detection feature and selected a different model for that particular shot), so I recreated it using Seedance 2.0 at 1080p. (I also tried Hailuo H3 and Kling 3.0, but in this particular case, Seedance 2.0 gave me the best overall result in one go.)
For the sound, I selected a solo-instrument soundtrack from Epidemic Sound and added extra sound effects to improve the overall sound design.
And here's the result.
I'm curious to hear what you think.
Did this bring back memories, or is this genuinely your first time hearing about WinRAR? Iâd love to know.
Will Bohn retweeted
Just FYI. If you are switching models in a session all the time - you are doing it wrong.
Every time you switch, your entire prompt cache is invalidated on the new model you switch to, and you have to repay the full input tokens price for all of it.
Stop doing this unless those models are free.
This is not a hermes thing - this is a fundamentals of inference thing.
Will Bohn retweeted
I turned 63. As I get older, the things I âhave to doâ and the things I âwant to doâ continue to increase. At the same time, the time I âhave to do themâ continues to decrease. I constantly need to reassess how I use my time and how I live my life. That said, once youâve taken something on, you canât simply put it down and walk away. This year also marks 40 years since I first started creating games, and 10 years since I founded Kojima Productions. From end of October through the end of the year, we are planning to hold the âKOJIMA PRODUCTIONS 10th Anniversary Exhibitionâ at Azabudai Hills Gallery. The Ludens at the entrance has also been changed to the â10th Anniversary Ludensâ version. While I continue to receive various opportunities and offers, I would like to remain focused on creating games.
Thank you for your continued support.
Will Bohn retweeted
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.
- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models
Tech Blog: z.ai/blog/glm-5.3
Will Bohn retweeted
One of the Keen researchers asked for guidance on compute expenditures. I replied: âVisualize the stack of $100 bills being set on fire with each run and weigh it against the knowledge sought.â In many ways, research today is in a better position to turn money into insight than ever before, but it can still be squandered.
While I really believe in all the potential with AI and everything (I've got my Skynet shirt ready); one of the biggest things that I give pause to is how all the cryptobros jumped over to the AI bandwagon faster than they created memecoins.
Just did a migration while using metalmonitoring.com - I'd say everything went fine!
This is a little premature to post, but because it's exciting (to me anyway), I wanted to share:
I'm building ConsensysPrime.com , an alternative to Usercentrics/Cookiebot, etc. etc. Yes, it's one of those wonderful things that pops up and asks if you want to have some cookies, and after you've had some cookies and shared them - what do you want to do with them -- things.
The reason: Everything out there in this area sucks and costs you an arm and a leg. Yet _all_ website/app owners are supposed to be compliant with GDPR and similar laws around the world. Are you set to handle Data Service Requests? Heck, most of my sites don't even have a "would you like a cookie?" popup because the solutions out there are expensive and don't do a whole lot (and I'm not collecting data...like I should be).
I've still got a lot of debugging, QA, and polish to do on it, but as someone who's passionate about marketing, I think this is an important project for the web, and for me.
I promised my brother that I'd have a little something to do website monitoring.
Enter: metalmonitoring.com
Not only does it monitor the sites, but if something is wrong (especially systematically), it uses AI to tell you what's up in plain English.
Now, I've just started with it, but you can expect that there to be more features, such as monitoring for changes - even on other people's webpages.
There's a free trial period - let me know what you think!
Hope you like where it's headed!
Will Bohn retweeted
New post from Joe Mullin, EFF's Sr. Policy Analyst: Publishers say theyâre blocking the Internet Archive because they fear AI scraping. This is a misguided notion: it wonât stop AI, but it will erase the webâs historical record.
@EFF has the story ⤾ď¸
eff.org/deeplinks/2026/03/blâŚ
#DigitalPreservation @WaybackMachine
Introducing Unsloth Studio â¨
A new open-source web UI to train and run LLMs.
⢠Run models locally on Mac, Windows, Linux
⢠Train 500+ models 2x faster with 70% less VRAM
⢠Supports GGUF, vision, audio, embedding models
⢠Auto-create datasets from PDF, CSV, DOCX
⢠Self-healing tool calling and code execution
⢠Compare models side by side + export to GGUF
GitHub: github.com/unslothai/unsloth
Blog and Guide: unsloth.ai/docs/new/studio
Available now on Hugging Face, NVIDIA, Docker and Colab.
I was only using CodeRabbit for their CLI access to review my junk code. They pulled it out of their monthly payments to a per-file-review "pay-as-you-go" pricing model. They tried to say it's "unlimited" and amazing in their blog post, but... no - I can see straight through that.
Bye-bye, CodeRabbit. đ°
coderabbit.ai/blog/introduciâŚ
