开启时会模糊预览图，关闭后正常显示
If you’ve never bumped into any of these people, try really really hard to build an accessible UI library and post about it online.
Creator of @tailwindcss. Listener of Slayer. Austin 3:16.
RT @USOPM: OPM’s new SEDP is about to shake up government leadership. From constitutional governance to crisis leadership, this program is…
Dir., Office of Personnel Management (previously, MP at a16z); Author of Secrets of Sand Hill Road; father of three amazing/crazy/beautiful girls.
There are two reasons that the billable hour has persisted as a mechanism for pricing legal services. The first, which is the most obvious, is that quality of service is in part a function of the experience of the attorneys doing the work and the amount of attention and time they spend. If lawyers were to offered a fixed price for their service, then unless it's a service consumed over and over so that the fixed price ends up being fair to both lawyer and client, it's a bad deal for one or the other. Otherwise lawyers bill by the hour (and raise their hourly fees until the market clears) because there's no way to scale their time and attention. But second and less obvious reason is because there are ethical rules against lawyers and clients sharing risk in the same business. This ethical, which has some historical roots in anti-immigrant (and anti-semitic) exclusionary practices by the bar over 100 years ago, has persisted in part because of its convenience to the elite top of the profession, which gets to ask what it wants for its services and isn't particularly concerned about accessibility to its services by new entrants. Silicon Valley law firms have gotten around this — to the mutual benefit of the law firms and their clients — by forming affiliates to accept equity investments in lieu of hourly fees. But by and large, billing by the hour remains the standard practice in the profession. With the advent of AI, the first reason for billing by the hour is falling away. But the second remains, and remains a problem for the profession. For legal services to become a truly scalable product, clients' and lawyers' interests must align through risk sharing. AI technology by itself isn't going to solve this problem.
Wonderer. Amor fati. Scaling trust.
OpenRouter 分析最近13个月的数据，总量超100万亿token。 发布的人工智能现状的报告。 对于理解当前的人工智能采用程度和发展，这份报告还是挺重要的。 ---------其中最重要的数据要点：------- 推理模型采用率大幅提高： 自OpenAI 发布推理模型 o1 后，使用从单次生成转向多步审议与推理； 目前经推理优化的模型承载的token占比已超过50%，成为默认路径。 闭源仍占多数，但开源持续增长： 2025年末约三分之一的使用为OSS。中国开源模型从1.2%周占比（2024末）攀升至某些周接近30%，全年周均约13.0%。 开源使用从深度集中过渡到多元竞争： 至2025年末，无单一模型长期超过25%的OSS份额，5–7个模型均衡分布。 “中等模型”成为新的“模型-市场匹配”主力： 按参数量划分：小于15B（小）、15–70B（中）、大于70B（大）。使用从“小”向“中”“大”转移。 编程增长迅猛： 编程成为增长最快类别，2025年初约11%，近期已超50%。开源模型的使用以角色扮演（≈52%）与编程为主；中国开源中角色扮演约33%，编程+技术合计约39%。 成功工具调用的请求占比全年上升： 平均prompt长度从~1.5K增至>6K（约四倍），completion从~150增至~400（近三倍）。平均序列长度从<2,000（2023末）到>5,400（2025末），编程类输入长度为整体的3–4倍，是主驱动。 大洲占比： 北美47.22%、亚洲28.61%、欧洲21.32%、南美1.21%、大洋洲1.18%、非洲0.46%。亚洲从~13%升至~31%。 国家Top10： 美国47.17%、新加坡9.21%、德国7.51%、中国6.01%、韩国2.88%、荷兰2.65%、英国2.52%、加拿大1.90%、日本1.77%、印度1.62%。 语言占比： 英语82.87%、简体中文4.95%、俄语2.47%、西班牙语1.43%、泰语1.03%、其他合计7.25%。 价格弹性低、市场未商品化： 质量/可靠性/集成度显著压倒价格。存在“杰文斯效应”迹象：高效低价模型引发更长上下文与更多迭代，推高总消耗。
详情：
Meituan is trying to compete with Z-Image in the «small but mighty» image generation/editing class. 6B DiT based on (you guessed it) Qwen 2.5 VL
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
Some have said they don't like the Ghostty project/non-profit because they believe it is a vanity project. It is not a vanity project, but instead of trying to convince you of that, I will instead share what I posted about how even if it was, it doesn't matter. A non-profit has a stated mission purpose and transparency. If the executive director or high-level employee of a nonprofit is an ego-inflated, vanity-soaked, attention-hungry narcissist, it doesn't matter unless it detracts from the purpose of the mission or causes some real, tangible harm. If they're achieving the goals of the mission, it doesn't matter. If they're not achieving the goals of the mission, then there is legal recourse to force them out. I've done a good amount of non-technical philanthropy the past few years and I have to say that in my experience a lot of people in that world are extremely high on their own supply. They're so drunk on their reality that they're the good-est possible people in the world and they're so virtuous and better than me or anyone else. And I don't care, I just want to know their mission, see their numbers, see their results, etc. and I'll donate. Of course, if you're personally disgusted by something like this in a way that you can't compartmentalize away from the broader mission of a project then not supporting it is fine. Not supporting a project for any personal reason is fine. I'm merely stating that they're separable topics. I don't think I'm that kind of person. And Ghostty is not a project made for vanity. I like to share my work online and I'm excited to talk about it. I don't think I talk about vacuous topics for the sake of attention. I'm not driven to do this for the attention it does or may get me. But, if you believe I do, so be it. (Note: this is a slightly edited copy of a Lobsters comment I posted)
Working on a new terminal: Ghostty. 👻 Prev: founded @HashiCorp. Created Vagrant, Terraform, Vault, and others. Vision Jet Pilot. 👨✈️