Worth reading.
Rockstar Games 周三解雇了我。第二天 GTA 6 就发售了
我是 Rockstar North 的高级引擎工程师,从事 RAGE 性能优化。三个月前,我在遥测模块里发现了一条本不该存在的数据管道,而且我能证明它流向了哪里
此刻每一份正在运行的 GTA 6 都在收集玩家的行为数据。这部分不算新闻。每款游戏都会收集遥测数据。隐私政策里写的是"用于改进未来产品的匿名游戏数据分析"。我发现的不是这个
遥测模块的数据会汇入 Take-Two 内部的一个系统,名为 Lens。Lens 并不会把匿名使用统计数据发送到 Rockstar 开发人员的数据看板。它为每一位玩家建立一份个人行为画像。不是匿名的。与你的主机账号、你的平台 ID、你的 IP 绑定
我在8月14日向我的技术主管报告了这件事。他说这是"用于玩家体验研究的扩展分析",让我别管这事。我要求提供数据处理协议。他说这是在 Take-Two 集团层面处理的,不在 Rockstar 的职责范围之内
我调取了 Lens 数据管道的完整文档。我发现的内容如下:
遥测的内容远远超出游戏时长和崩溃报告。Lens 以毫秒级精度记录每一次输入。按键时间点、摇杆动作、反应速度、PS5 自适应扳机上的扳机压力模式。它实时运行行为分析,并生成他们称之为"玩家行为签名"的东西 - 一份对每位玩家都独一无二的画像,就像指纹对每个人都独一无二一样
Lens 会从你的游戏过程中推导出:冲动控制评分、风险承受度指数、攻击性反应模式、耐心阈值、压力下的决策速度、消费冲动评级。这些不是调试指标。这些是心理行为类别
直到我找到了对外传输的管道,我才明白它们是用来做什么的
这些数据不会留在 Rockstar。也不会留在 Take-Two。Lens 通过内部 API 将行为画像传送给 Take-Two 旗下一家名为 Vantage Analytics 的子公司。Vantage 没有公开网站,没有新闻页面,没有招聘信息。它于2023年在特拉华州注册成立
Vantage 是一家行为 AI 公司。它用玩家画像训练预测模型,并把这些模型授权给商业客户。管道文档中包含一份不完整的客户清单。两家保险集团,一家信用风险建模公司,一个人力资源筛选平台,以及一家广告数据经纪商
你的 GTA 6 游戏过程此刻正在生成一份行为评分。没人追你的时候,你开得有多快。你会不会在红灯前停下。NPC 无端攻击你时,你怎么反应。你行动之前要想多久。你花游戏内的钱有多冲动。每一个选择都在喂养一个模型,企业正在付费查询这个模型,用来对真实的人做决定
63
I ranked 8 growth tactics by what I’d test first.
The results come from studies + case reports.
The ranking is mine.
48
Over the years, I figured the edge in any startups is all about how quickly you can recover from rejection.
This means:
- Hearing NO hundreds of times and still making another call before getting a YES
- Losing a deal and figuring out why vs. blaming the guy on the phone
- Launching something nobody buys (yes, we all been there) + changing the offer instead of adding more features
- Being able to get brutal feedback with a straight face + using it instead of defending your baby (aka your product)
- Watching some random guy outperform you and studying what he did instead of crying about it
Now me personally, I never was immune to rejection, but I learned how to turn it unto info can use to improve
I also learned many fast NO's= faster YES's
I dug up the old homepages of ten AI companies that hit $1M+ ARR.
Turns out, ALL of them pivoted their landers multiple times.
Here’s what they used to sell 👇
1
64
Easy:
$20k offer that saves a company $100k
Hard:
$2k offer that promises to change somebody's life
Even Harder:
$9/month offer in a massive niche with a free trial
---------------
*The lower your price, the thinner your margin for error on CAC (customer acquisition cost)
93
39 overlooked industries and niches that could pay $15–25K for AI automation services:
#1. Importers: folks manually chase customs docs, supplier payment/ shipping updates, etc
#2. Medical labs: staff coordinates samples, test results + comms with doctors
#3. Private security companies: managers manually juggle last-minute shifts and incident reports
#4. Corporate event planners: teams chase vendors, contracts, quotes and updates
#5. Solar installation companies: think manually managing site inspections, permits + installations
#6. Restaurant franchises: managers manually track inventory, orders and issues across locations
#7. Pest control companies: staff schedule recurring visits/ renewals/ prep compliance certificates
#8. Car dealerships: folks manually follow up with leads and vehicle paperwork
#9. Catering companies: employees coordinate guest counts/ menu changes, suppliers + delivery schedules
#10. ISO certification services: employees chase people for compliance docs + audit evidence + corrective actions
#11. Pharmaceutical distributors: folks manually track batches, expiration dates, orders + a bunch of paperwork.
#12. Waste management companies: dispatchers coordinate pickups, certificates/ regulatory reports.
#13. Private universities: admins sort out student applications, docs and tuition payments
#14. Elevator maintenance companies: folks track mandatory inspections/ service visits + certificates
#15. Film production companies: producers coordinate locations/ permits/ budgets/ shooting schedules
#16. Veterinary chains: docs manually handle vaccination reminders + recurring treatments
#17. Corporate travel agencies: agents manage itinerary changes/ travel approvals + billing.
#18. Regional telecom providers: ops teams do installs, service complaints + tech visits
#19. Parking garage operators: staff manage monthly subscriptions/ payments + incidents and complaints
#20. Big landscaping companies: managers coordinate recurring visits, crews + customer reports
#21. Cold storage operators: employees manually monitor temperature logs + stock movements + compliance records
#22. Industrial inspection companies: inspectors juggle site visits/ evidence collection/ reports + certification deadlines.
#23. Clinical research orgs: teams track trial docs, site comms/ regulatory deadlines
#24. Corporate relocation companies: also teams coordinate inventories/ moving crews, quotations/ customer claims
#25. Agricultural machinery dealers: staff manages spare parts/ warranty claims/ maintenance
#26. Corporate uniform suppliers: folks manually collect employee sizes/ manage replacements + bulk deliveries
#27. Industrial laundries: teams track garment collection (and a ton of missing items) + deliveries + customer billing.
#28. Fleet management companies: Teams track vehicle docs + fines + insurance renewals + maintenance schedules.
#29. Shipping container rental companies: teams manage availability/ pickups + returns + rental charges.
#30. Fire safety maintenance companies: techs track extinguisher inspections + service deadlines/ compliance certificates
#31. B2B electronics repair companies: teams manually track repair tickets/ spare parts + customer approvals
#32. Marina operators: folks coordinate boat berths + maintenance services/ contracts/ renewals
#33. Energy auditing companies: Consultants collect utility data, schedule inspections and prepare repetitive reports.
#34. Vending machine operators: folks manually coordinate restocking/ machine failures + route planning/ inventory
#35. Portable toilet rental companies: dispatchers have to look after cleaning routes + equipment availability, pickups/ contracts
#36. Insurance claims administration companies: folks chase docs, coordinate adjusters/ prepare claim reports
#37. Document storage companies: Employees manually process file retrieval requests, pickups and document tracking.
#38. Equipment calibration companies: folks schedule inspections/ prepare certificates + monitor calibration deadlines
#39. Condo management companies: staff handle a ton resident complaints + maintenance requests + contractor follow-ups/ unpaid fees
The main thing businesses that “need AI” is those where expensive employees spend hours every week chasing people/ docs/ info/ manage sustems.
Say a company spends $100k/ year doing manual process.
The you can introduce automation and remove 80% of that part of that cost for a raction of the cost.
That’s the oppprtunity.
1
1
120
I know enough direct response marketers selling mediocre SaaS and still killing it.
I also happen to know hundreds of tech founders who know how to build amazing products and sell shit.
The difference is marketers start with things people are willing to pay for, and builders do what they THINK people would buy.
This is because builders love dreaming + building. It's their passion.
Marketers love dreaming too... but they actually go out there and throw their ideas in front of people, even before spending months building something.
That's why ok-ish products make millions (because the code might be full of bugs but the actual features have real value)
So, if you're a technical founder, for every 8 hours you go building, spend at least 1 hour talking people.
You'll be surprised with results.
Mistakes I'd avoid while automating with AI:
- Spending weeks building automations nobody will use
- Buying shiny AI tools before figuring the actual problem
- Building complex AI agents when a simple workflow would do
- Not measuring how much time/money automation actually saves
- Think automation comes without maintenance
- Trying to automate everything (aka forgeting the 80/20 rules)
- Autoamating where mistakes are expensive
So, even before talking to Codex/Claude, ask yourself this:
1. How often does this problem happen?
2. How much $$$ does it cost?
3. What happens if AI slop hits the fan?
If you can't answer all of the 3, you're not ready to automate.
6
95
Is Codex getting a new Astra or Sol, or shall we consider this battle lost..?
Just a tiny reminder that the older generation of Opus actually beats OpenAI at coding (terminal bench 4.0 test):
- Claude Opus 5.5: 64.8%
- GPT 6 Astra: 58.2%
- GPT 6.1 Sol: 58.2%
This is 2 OpenAI models delivering the exact same score while Claude is 6.6 points ahead!
At this point OpenAI doesn't need another 200-page model announcement.
We just need to see something that actually writes better code.
CC: @thsottiaux
3
100
Most personal brands are disguised distribution channels pretending to be a real biz.
You can have 100K followers and still don't get a dime in return.
Having followers only means you have attention.
And those eyeballs is great ONLY if you actually do something with it.
Simple math:
100,000 cat meme lovers can't afford a $5k offer, but 2,000 entrepreneurs or high wealth followers can.
Which audience would you have?
My vote is high wealth followers... but still a gazillion people think all likes are equal.
Twitter is a great example.
People think follower count is more important that follower quality.
Big mistake.
Followers are not customers and likes are not $$$.
And if you're a founder, going viral doesn't automagically mean sales.
Right audience + quality offer makes sales. Rest is noise.
1
98
Please people, it's time to realize nobody actually wants to buy AI automation!
They just want to stop worrying about the problem you solve.
This how seduction works btw.
Most folks think seduction= people falling in love with you
But in reality selling/ seduction= making fall in love with themselves.
That's why the best sales pitch is when they start selling themselves on your product.
4
187
Your last 20 lost sales calls have more useful growrh tips than all AI marketing prompts.
Just export your last lost deals from HubSpot/Salesforce (or whatever CRM you're using).
If you’re not using a CRM, just make sure to get the transcriptions.
Now give that data to Claude/ChatGPT and ask it to challenge yourself by being the devils’s advocate.
Look for:
- The non obvious stuff for losing the deal
- Main objections and the way you handled them in wrong/right way
- Hidden language patterns you didn’t recognize
For example, let’s say your customer says your offer is "too expensive"
Was it that you never helped the customer figure the cost of the problem? (they didn’t see the value)
Was it they didn’t believe this would work?
Or did the alternatives sound better in their mind?
Once you have multiple objections, rank the problems by how often they appear.
In my case, we never attack all at once. I better focus on the main 3 things and get it into a feedback loop.
For example:
Not asking enough questions about customers pain usually make you lose sales because they can’t compare it to the outcome you’re promising during the call.
AI can tell you that and help ypou insert the right questions in the next calls.
And btw, don't ask AI what features you should build.
Ask it what you can fix without adding new stuff.
1
82
Denis Balitskiy retweeted
reminder
if you think your life is hard, this was Charlie Munger's reality at age 31
divorced, dead broke, and having just buried his 9-year old son who died of leukemia
warrior spirit to even RECOVER from that, much less have a life that turned out like his...
My old life use to call me when I was afraid to move on. Usually it happened because I wasn't certain where I wanted to go.
And that's how I would drag mistakes for years.
But some time ago something changed.
I stopped dreaming about this "new and perfect life" and started asking myself what I wouldn't accept as personal standards.
Sounds like a small difference, but it changed a lot of things for me.
When you're clearing the things you don't want to see anymore, you can choose a new direction more strategically.
Also, you become more careful about the decisions you make.
In my case, the old Denis used to chase opportunities/ confusing being busy with working.
It would be something like this:
Any given day, when I had a personal issue, I'd get into this ruminating state for hours.
This would go on and no for many years.
But later, I've learned to identify those moments + made unacceptable for me to do that.
Again simple rule, but big change.
So don't accept the old you that drags you. It's just a ghost trying to fuck your life.
1
99


















