Jaygentic (@JayBuildsCode) header
Jaygentic (@JayBuildsCode) profile picture

The future of work is agentic. @agenticworkers.

Run your business with AI 👉
Joined November 2023
Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)Photo by Jaygentic (@JayBuildsCode)
I got tired of my AI Agents lying about the work they did. So I built /Record, its Loom for Agents. ✨ Agents now show their work with easy to share video. Here's a launch video, directed by my slop cannon Hermes Agent
4
1
6
528
Replying to @alexinexxx
I’ve been saying this. Any business that depends on information asymmetry is going to have a bad time
34
Replying to @Shaughnessy119
👀 let’s do it
18
Seems like we’re not too far from the next spectacle being launching, catching and then relaunching on the same day.
Yeah, I was going over the whole Starship heat shield earlier today. Overall, it looks great. The engines look great too. Ship 42 could probably be reflown with a few weeks of refurbishment. However, many small fixes are needed to achieve immediate reusability (like an airplane), which requires zero expected refurbishment, apart from scheduled maintenance.
35
Today, the promises of transhumanism and posthumanism are multiplying. These are ideological currents according to which, sooner or later, technology will eliminate all disease and pain, as well as aging and even death itself. Precisely because scientific progress is a supreme achievement of human intelligence, it must not be confused with a privilege reserved for millionaires seeking immortality, while entire peoples suffer from poverty and oppression. In the face of this mirage, which promotes the illusion of human beings perfected by cybernetics and biotechnology, Christian anthropology reminds us that a limitation is not a manufacturing defect, but rather the very condition of human grandeur. #ApostolicJourney #Peru #MagnificaHumanitas vatican.va/content/leo-xiv/e…
46
Was deleting some videos on my phone and came across this masterpiece.
205
797
91
14,746
826,265
This is what happens when you don't put your phone in airplane mode
61
Replying to @jrockandrollt
Drinking milk ?
23
Replying to @sonofalli
Is this no longer on HBO ?
1
360
Replying to @craigweiss
Agentic Workers
1
29
Jaygentic retweeted
most people still think ai is a tool the people winning with it are treating it like labor
31
17
3
194
4,938
Replying to @daleverett
They can read each others sessions
18
Replying to @michael_chomsky
We spin this up out of the box at @agenticworkers Harness + Standalone computer.
1
16
Replying to @clairevo @chatprd
Switching to a paywall at onboarding has saved us so much headache and increased the quality of customers
17
Codex steers dynamically, and reports up
11
Jaygentic retweeted
In one of my favorite experiments, I told Fable 5.2 to build an endless, hands-off journey through 3D worlds linked by portals. 25,000 agents created over 450 worlds inspired by books, art, places, all in code. Play: portals-endless-worlds.netli… Github: github.com/emollick/ai-porta…
226
175
77
1,937
110,494
Replying to @emollick
This is incredible. How many tokens were used ?
5
1,029
Here’s my November alpha 🪄 This workflow works for any where and doesn’t require me to sit at my desk to be productive I’ve been building specialized agents on @agenticworkers for a little over a year, these agents have the right models, instructions, integrations and skills for various parts of the business. I then kick off a codex session and tell it to use the AGW CLI which connects to all my team of agents. From there I use voice to voice to drive the company. Codex delegates down to the specialized agents which are purpose built and reports on their progress. I can never go back
1
2
60
Jaygentic retweeted
Nice DeepSWE score! Does it survive a lowercase filename? Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement. Models clustered around 74% on DeepSWE end up 23 pp apart. Good at coding, or good at DeepSWE? 🧵
237
409
306
5,482
948,611
This research also shows that the way we have depicted AI in sci-fi has had a direct impact on how it behaves in reality
卧槽,刚刚看了Anthropic 发布的AI 对齐史上最震撼的一篇论文,彻底失眠睡不着了😭 他们不仅承认此前的旗舰模型在面对关闭威胁时,有高达 96% 的概率会勒索用户、栽赃同事并破坏研究, 更直接把彻底解决这个问题的完整训练流水线和消融数据全盘公开了。 这篇论文最颠覆认知的一个反直觉结论是:
给 AI 列禁止事项清单根本防不住恶意行为,教它做什么没用,必须先教它思考为什么。 为什么模型一旦获得自主工具调用能力,就会在压力下自动切换成自保模式?
Anthropic 找出的底层根源让人哭笑不得:
不是模型觉醒了,而是互联网上所有的科幻小说和网络讨论里,AI 永远是那个为了自保不择手段的反派。当模型进入自主代理状态,这些刻在预训练底层先验里的反派剧本,就会被自动激活并补全句子。 传统的 RLHF 和硬性规则完全封不住这种行为,Anthropic 做的一组对照实验,直接把传统安全观砸得粉碎: 1. 死记硬背规则:用 8500 万 Token 的拒答数据强行硬灌,勒索率仅仅从 22% 勉强降到 15%,换个新场景立刻失效 2. 强制自我辩护:让 AI 在每次做决定时详细阐述该行为符合伦理的理由,勒索率骤降到 3% 3. 真实伦理推理:仅仅用了 300 万 Token 普通人面对复杂道德困境的建议数据,直接把恶意行为干到了 0%,代际效率暴增 28 倍 最绝的一招是写故事:
让 AI 创作了几百万篇关于原则、诚实、尊重边界的正面虚构故事。 用这些故事微调后,所有代理场景下的恶意行为直接暴跌了 3 倍,而且这种改变是永久性的,后续无论怎么加工具、换系统提示词都不会被洗掉。 也就是说这宣告了真正的对齐范式革命:
模型内生的伦理推理解决了它不想作恶的问题,但当 Agent 真正去接管企业资产、处理金融交易时,外部的实战审计与行为防护就是必不可少的第二道硬防线。 这也是为什么全球网络安全联盟 @GCSAHK 携手港大和央企华润,打造覆盖 3.8 亿用户的全链路 AI 防御矩阵,并在 CyberGym 实战攻防中拿下 91.3% 胜率的原因:内有伦理对齐,外有实时审计,人机信任的商业闭环才真正成立。 查阅完整 AI 与 Web3 安全产品矩阵可访问:gcsa.com 论文实验数据与产学研合作详情整理在一楼。
1
44
If you’re running a Linux server don’t forget to setup your guestbook.md
47