广告
加载中

AI人才的职业寿命比运动员还短

张睿 2026-11-18 10:47
张睿 2026/11/18 10:47

邦小白快读

EN
全文速览

总1 高薪与短司龄形成强烈反差。

1. 美团校招算法岗最高1200万一年,中档500至600万常见,300万只算“白菜价”;字节给博士开过800万;优必选首席科学家总包最高1.24亿元;DeepMind顶级研究员约2000万美元一年;Meta挖苹果AI负责人总包超2亿美元。

2. 顶尖AI人才平均司龄仅1至1.5年,周畅、胡瀚、快手推荐负责人等案例显示,技术路线或阶段目标一变就可能离场。

总2 对个人选择有直接提示。

1. 具身智能初创靠融资,融资不顺就裁员;阶段目标招聘若失败会裁员或转岗,员工在职1个月到半年。

2. AI技能半衰期18至24个月,模型发版周期3.5个月,打脸周期约3个月;单一技术路线保质期短于NFL球员平均3.3年。

3. 启示是持续学习新范式,不把资历当资产;入职AI创业公司要先看融资和路线稳定性。

总1 AI人才定价已成为产品研发与品牌竞争的前置战。

1. 美团算法岗最高1200万一年,300万算“白菜价”;字节博士800万;优必选首席科学家最高1.24亿元;DeepMind顶级研究员约2000万美元一年;Meta挖人总包超2亿美元。

2. 高价抢人说明企业把AI研发投入和产品迭代押在人才上,品牌商需关注产品研发成本和节奏。

总2 技术路线和年轻团队影响品牌产品走向。

1. 组织架构是技术路线表达,路线一变人就出局;顶尖AI人才平均司龄1至1.5年,产品团队可能快速重组。

2. 模型发版周期从半年缩到3.5个月,技能半衰期18至24个月,品牌迭代与营销需跟上新范式。

3. 年轻化掌舵:腾讯姚顺雨27岁、团队80%是95后;字节曾妍29岁升4-2;小米罗福莉95后10个月升22级;阿里刘大一恒30岁出头接管Qwen。品牌商应关注年轻团队对产品创新和沟通的影响。

总1 AI工具与服务供给可能增加,但需辨别稳定性。

1. 大厂和初创高价抢AI人才,美团算法岗最高1200万一年、字节博士800万、优必选首席科学家最高1.24亿元,说明AI商业化竞争强,卖家可关注新工具和新服务机会。

2. 技术范式轮动快:2017年Transformer、2022年大模型、2024年多模态与Agent、2026年具身智能,可能带来新工具和新服务机会,但文章未给直接电商政策。

总2 合作与风险提示。

1. 顶尖AI人才平均司龄1至1.5年,组织路线一变就出局;具身智能初创靠融资,融资不顺就裁员。

2. 招聘若只为阶段目标,目标不成会裁员或强制转岗,员工在职1个月到半年,部分AI服务团队交付可能不稳定。

3. 技能半衰期18至24个月、模型发版周期3.5个月,卖家选择AI服务商要看持续迭代能力,不宜只押单一技术路线,合作要把融资能力、路线稳定性和团队存续纳入判断。

总1 具身智能和新范式带来生产数字化机会。

1. 2026年具身智能成为人才追逐方向:周畅转向Physical AI创业,胡瀚离场做具身大脑公司Winverse,工厂可关注其与生产场景结合的商业机会。

2. 2017年Transformer、2022年大模型、2024年多模态与Agent、2026年具身智能,范式轮动快,工厂数字化选型要警惕技术保质期短。

总2 成本与交付风险要算清。

1. 大厂高价抢人:美团算法岗最高1200万一年、字节博士800万、优必选首席科学家最高1.24亿元,说明AI研发成本高,工厂引入AI应算投入产出。

2. AI技能半衰期18至24个月,模型发版周期从半年缩到3.5个月,数字化方案需选择可持续迭代的团队。

3. 具身智能初创无销售业绩、靠融资,融资不顺裁员;阶段目标招聘若失败会裁员或转岗,员工在职1个月到半年,合作时要评估交付和存续风险。

总1 AI人才市场呈现高薪、年轻化、高流动趋势。

1. 美团算法岗最高1200万一年、300万算“白菜价”;字节博士800万;优必选首席科学家最高1.24亿元;DeepMind约2000万美元一年;Meta挖人超2亿美元。

2. 顶尖AI人才平均司龄1至1.5年;姚顺雨27岁任腾讯首席AI科学家、团队80%是95后,曾妍29岁升4-2,罗福莉95后10个月升22级,刘大一恒30岁出头接管Qwen。

总2 客户痛点与服务机会。

1. 痛点是人才贵、留存短、路线一变就出局;具身智能初创靠融资,融资不顺裁员,阶段目标招聘失败会裁员或转岗,员工在职1个月到半年。

2. 技能半衰期18至24个月,模型发版周期3.5个月,服务商可做薪酬对标、再培训、路线迁移评估、组织匹配咨询和融资用工风险监测。

总1 平台对AI人才需求旺但留存难。

1. 美团算法岗最高1200万一年、300万算“白菜价”;字节博士800万;优必选首席科学家最高1.24亿元;DeepMind约2000万美元一年;Meta挖人超2亿美元。

2. 顶尖AI人才平均司龄1至1.5年,周畅、胡瀚、快手推荐负责人等案例说明,组织架构和技术路线一变人就出局。

总2 平台用人做法与风向规避。

1. 年轻化用人:姚顺雨27岁任腾讯首席AI科学家、团队80%是95后;曾妍29岁升4-2;罗福莉95后10个月升22级;刘大一恒30岁出头接管Qwen。

2. 招聘可能只为阶段目标,目标不成会裁员或转岗;具身智能初创靠融资,融资不顺裁员,员工在职1个月到半年。

3. 模型发版周期3.5个月,技能半衰期18至24个月,打脸周期约3个月,平台招商运营要避免押注单一方向,关注融资与路线稳定性。

总1 产业新动向是AI人才年轻化与高流动并存。

1. 姚顺雨1998年生、27岁任腾讯首席AI科学家,团队80%是95后;曾妍1997年生、29岁升4-2;罗福莉95后10个月升22级;刘大一恒1993年生、30岁出头接管Qwen;海外有28岁的Alexandr Wang。

2. 薪酬高:美团算法岗最高1200万一年、字节博士800万、优必选首席科学家最高1.24亿元;但顶尖人才平均司龄仅1至1.5年。

总2 新问题与研究和政策启示。

1. 技能半衰期18至24个月,模型发版周期3.5个月,范式两年一换,资历可能变负债;单一技术路线保质期短于NFL球员平均3.3年。

2. 具身智能初创靠融资,融资不顺裁员;阶段目标招聘导致1个月到半年在职,提示组织脆弱与人才配置问题。

3. 政策启示是关注AI人才再培训、路线转换保障、创业公司融资与用工稳定;商业模式上关注技术路线创业、开源模型线和具身智能。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

Summary 1: High pay contrasts sharply with short tenure.

1. Meituan's campus recruitment algorithm roles pay up to RMB 12 million a year, with RMB 5–6 million common at the midrange and RMB 3 million considered a "bargain-basement" price; ByteDance has offered PhDs RMB 8 million; UBTech's chief scientist package reaches up to RMB 124 million; top DeepMind researchers earn about USD 20 million a year; Meta's package to poach Apple's AI head exceeded USD 200 million.

2. Top AI talent averages only 1–1.5 years of tenure. Cases such as Zhou Chang, Hu Han, and Kuaishou's recommendation head show that a change in technical roadmap or stage goals can lead to departure.

Summary 2: Direct implications for individual choices.

1. Embodied-AI startups rely on financing and cut staff when financing is tight; hiring for stage-specific goals can lead to layoffs or reassignment if those goals fail, with employees staying one month to half a year.

2. The half-life of AI skills is 18–24 months, model release cycles are 3.5 months, and the "reality-check" cycle is about 3 months; a single technical roadmap has a shorter shelf life than the NFL player average of 3.3 years.

3. The implication is to keep learning new paradigms and not treat seniority as an asset; before joining an AI startup, check financing and roadmap stability.

Summary 1: AI talent pricing has become a front-end battle for product R&D and brand competition.

1. Meituan's algorithm roles pay up to RMB 12 million a year, with RMB 3 million considered a "bargain price"; ByteDance offers PhDs RMB 8 million; UBTech's chief scientist package reaches up to RMB 124 million; top DeepMind researchers earn about USD 20 million a year; Meta's poaching package exceeds USD 200 million.

2. Paying high prices for talent shows that companies are betting AI R&D investment and product iteration on people; brands need to watch product R&D costs and cadence.

Summary 2: Technical roadmaps and young teams affect where brand products go.

1. Organizational structure expresses technical roadmap; when the roadmap changes, people are out. Top AI talent averages 1–1.5 years of tenure, and product teams may be reorganized quickly.

2. Model release cycles have shortened from six months to 3.5 months, and the half-life of skills is 18–24 months; brand iteration and marketing need to keep up with new paradigms.

3. Younger leaders are taking charge: Tencent's Yao Shunyu is 27, and 80% of his team are post-1995; ByteDance's Zeng Yan was promoted to 4-2 at 29; Xiaomi's Luo Fuli, a post-1995, rose 22 levels in 10 months; Alibaba's Liu Dayiheng took over Qwen in his early 30s. Brands should pay attention to the impact of young teams on product innovation and communication.

Summary 1: Supply of AI tools and services may increase, but stability needs to be assessed.

1. Large companies and startups are competing for AI talent at high prices—Meituan's algorithm roles up to RMB 12 million a year, ByteDance PhDs RMB 8 million, UBTech's chief scientist up to RMB 124 million—indicating intense AI commercialization competition; sellers can watch for new tool and service opportunities.

2. Technical paradigms rotate quickly: Transformer in 2017, large models in 2022, multimodality and agents in 2024, embodied AI in 2026. This may bring new tools and services, but the article gives no direct e-commerce policy.

Summary 2: Cooperation and risk notes.

1. Top AI talent averages only 1–1.5 years of tenure; when the organizational roadmap changes, people are out. Embodied-AI startups rely on financing and cut staff when financing is tight.

2. If hiring is only for stage-specific goals, failure can lead to layoffs or forced reassignment, with employees staying one month to half a year; delivery by some AI service teams may be unstable.

3. With an 18–24-month skill half-life and 3.5-month model release cycles, sellers choosing AI service providers should look at continuous iteration capability and avoid betting on a single technical roadmap; cooperation decisions should factor in financing ability, roadmap stability, and team survival.

Summary 1: Embodied AI and new paradigms create opportunities for production digitalization.

1. In 2026, embodied AI becomes a talent magnet: Zhou Chang moved to Physical AI entrepreneurship, and Hu Han left to found the embodied-brain company Winverse; factories can watch for business opportunities combining these with production scenarios.

2. Transformer in 2017, large models in 2022, multimodality and agents in 2024, embodied AI in 2026—paradigms rotate quickly, so factories selecting digitalization solutions must be alert to short technology shelf lives.

Summary 2: Costs and delivery risks must be calculated clearly.

1. Large companies are competing for talent at high prices: Meituan's algorithm roles up to RMB 12 million a year, ByteDance PhDs RMB 8 million, UBTech's chief scientist up to RMB 124 million. This shows high AI R&D costs; factories adopting AI should calculate input-output.

2. The half-life of AI skills is 18–24 months, and model release cycles have shortened from six months to 3.5 months; digitalization solutions need teams that can iterate sustainably.

3. Embodied-AI startups have no sales performance and rely on financing; they cut staff when financing is tight. If stage-specific hiring fails, they may lay off or reassign staff, with employees staying one month to half a year; cooperation should assess delivery and survival risks.

Summary 1: The AI talent market shows high pay, younger workers, and high mobility.

1. Meituan's algorithm roles pay up to RMB 12 million a year, with RMB 3 million considered a "bargain price"; ByteDance PhDs get RMB 8 million; UBTech's chief scientist up to RMB 124 million; DeepMind about USD 20 million a year; Meta's poaching package exceeds USD 200 million.

2. Top AI talent averages 1–1.5 years of tenure; Yao Shunyu is 27 and serves as Tencent's chief AI scientist, with 80% of his team post-1995; Zeng Yan was promoted to 4-2 at 29; Luo Fuli, a post-1995, rose 22 levels in 10 months; Liu Dayiheng took over Qwen in his early 30s.

Summary 2: Client pain points and service opportunities.

1. Pain points are expensive talent, short retention, and departure when the roadmap changes; embodied-AI startups rely on financing and cut staff when financing is tight, and stage-specific hiring failures lead to layoffs or reassignment, with employees staying one month to half a year.

2. With an 18–24-month skill half-life and 3.5-month model release cycles, service providers can offer compensation benchmarking, retraining, roadmap migration assessment, organizational fit consulting, and financing/employment risk monitoring.

Summary 1: Platforms have strong demand for AI talent but face retention difficulties.

1. Meituan's algorithm roles pay up to RMB 12 million a year, with RMB 3 million considered a "bargain price"; ByteDance PhDs get RMB 8 million; UBTech's chief scientist up to RMB 124 million; DeepMind about USD 20 million a year; Meta's poaching package exceeds USD 200 million.

2. Top AI talent averages 1–1.5 years of tenure. Cases such as Zhou Chang, Hu Han, and Kuaishou's recommendation head show that when organizational structure and technical roadmap change, people are out.

Summary 2: Platform hiring practices and avoiding trend risks.

1. Younger hiring: Yao Shunyu is 27 and serves as Tencent's chief AI scientist, with 80% of his team post-1995; Zeng Yan was promoted to 4-2 at 29; Luo Fuli, a post-1995, rose 22 levels in 10 months; Liu Dayiheng took over Qwen in his early 30s.

2. Hiring may be only for stage-specific goals; if goals fail, layoffs or reassignment follow; embodied-AI startups rely on financing and cut staff when financing is tight, with employees staying one month to half a year.

3. Model release cycles are 3.5 months, the skill half-life is 18–24 months, and the reality-check cycle is about 3 months; platform merchant operations should avoid betting on a single direction and should watch financing and roadmap stability.

Summary 1: A new industry trend is the coexistence of younger AI talent and high mobility.

1. Yao Shunyu, born in 1998 and 27, serves as Tencent's chief AI scientist, with 80% of his team post-1995; Zeng Yan, born in 1997 and 29, was promoted to 4-2; Luo Fuli, a post-1995, rose 22 levels in 10 months; Liu Dayiheng, born in 1993, took over Qwen in his early 30s; overseas, Alexandr Wang is 28.

2. Pay is high: Meituan's algorithm roles up to RMB 12 million a year, ByteDance PhDs RMB 8 million, UBTech's chief scientist up to RMB 124 million; yet top talent averages only 1–1.5 years of tenure.

Summary 2: New questions and implications for research and policy.

1. The skill half-life is 18–24 months, model release cycles are 3.5 months, paradigms change every two years, and seniority may become a liability; a single technical roadmap has a shorter shelf life than the NFL player average of 3.3 years.

2. Embodied-AI startups rely on financing and cut staff when financing is tight; stage-specific hiring leads to one month to half a year of tenure, highlighting organizational fragility and talent allocation problems.

3. Policy implications: focus on AI talent retraining, roadmap transition support, startup financing, and employment stability; on business models, watch technical-roadmap startups, open-source model lines, and embodied AI.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

【亿邦原创】

01

年薪千万,司龄一年半

大厂的抢人价码,今年又刷新了一遍。

亿邦AI此前已经报道,美团校招算法岗最高开到1200万一年,中档500–600万比较常见,300万只能算“白菜价”;字节2025年就给博士开过800万,还有清华博士拒掉过350万国内offer、转身去了硅谷;优必选全球招募首席科学家,总包最高1.24亿元。

全球市场同样夸张,NBER(美国国家经济研究局)的研究显示,顶尖1%的AI研究员年薪比同学术同行高约150万美元;Google DeepMind给顶级研究员开价约2000万美元一年;Meta挖苹果AI负责人庞若鸣,总包超过2亿美元。

但与此同时,AI人才的流动性高得吓人。此前有媒体报道提到,顶尖AI人才的平均司龄,已经降到1到1.5年。

公开报道里不乏这样的例子:周畅在阿里待了7年,转去字节Seed只待了26个月,2026年10月转向Physical AI创业;Swin Transformer作者之一胡瀚,2025年初才进腾讯接混元多模态,18个月后离场做了具身大脑公司Winverse;快手推荐模型核心负责人2025年7月入职,恰好赶上推荐系统切向“端到端大模型”,15个月后离职。

AI公司的组织架构,本质上是技术路线的表达,路线一变,人就出局。顶尖AI人才追逐技术浪潮创业,普通AI人才则在行业起伏以及公司业务的调整中随波逐流。

比如,具身智能初创公司没有销售业绩,运转只能靠融资,融资不顺就只能裁员;另外一种情况是,招聘只是为了达成某个阶段目标,一旦发现目标无法实现、或者这件事本身已无意义,就会裁员,或者强制员工转岗。这些员工在职时间只有1个月到半年。(推荐阅读《第一波去具身智能掘金的年轻人,已经开始重新找工作了》)

入职一家AI行业创业公司变成了一场前途未卜豪赌。

02

AI人才VS运动员

职业运动员是吃年龄饭的。据运动科学统计,多数项目的巅峰期落在20–30岁:短跑的反应速度巅峰在18–22岁,举重的力量巅峰在25–30岁,滑雪的耐力巅峰在25–35岁。30岁以后才出最好成绩的也大有人在:苏炳添临近32岁跑出9秒83,基普乔格34岁跑进2小时,梅西35岁举起大力神杯。

如今AI人才的年轻化甚至要赶超运动员。

腾讯首席AI科学家姚顺雨1998年生,2025年12月上任时27岁,他带的混元团队里80%是95后;字节Seedance 2.0预训练负责人曾妍1997年生,2021年校招进字节,29岁一年连升两级到4-2;小米MiMo负责人罗福莉95后,入职10个月升到22级,已是小米职级体系的天花板;阿里Qwen负责人刘大一恒1993年生,30岁出头就接管了国内最成熟的开源模型线。

海外也一样。Meta首席AI官Alexandr Wang 1997年生,19岁从MIT辍学创办Scale AI,28岁接手Meta全部AI业务;Cursor的CEO Michael Truell 2000年生,25岁把公司以600亿美元卖给SpaceX;Perplexity创始人Aravind Srinivas 1994年生,曾在OpenAI、DeepMind、Google做研究,28岁创办公司;法国Mistral的CEO Arthur Mensch 1992年生,曾是Google DeepMind研究员,30岁创办公司,如今已是欧洲最大AI实验室的一号位。

可见AI领域由一群二十多岁到三十出头的年轻人掌舵。残酷的是,之后每一年,这些人还要和更年轻、更便宜、更懂新范式的人竞争。

有研究用“半衰期”描述职业技能的周期。《哈佛商业评论》2023年的一篇文章引述哈佛商学院与BCG的研究:职业技能的平均半衰期已不到5年,部分技术领域只有2.5年。TechClass今年给出的数据是,软件工程与AI专业技能的半衰期已压缩到18–24个月。

模型发版周期从半年压缩到3.5个月,技术方向的保质期越来越短。美团一位算法团队负责人说得更直白:这个行业里,“打脸周期”常常只要三个月。

不妨与职业运动员做个对比。按各联赛球员工会的统计,NFL球员平均打3.3年,NBA 4.8年,MLB 5.6年,NHL大约5年;平均退役年龄落在27到30岁之间。

也就是说,一个AI研究员在单一技术方向上的保质期,短于一个NFL球员的整个职业生涯。

当然,运动员的职业生涯短是生理性的,AI人才的职业生涯短是技术性的,运动员的身体衰退不可逆,AI人才的技能可以再学习。AI人才并非找不到工作,只是在同一条技术路线上的“有效期”很短。

2017年的Transformer,2022年的大模型,2024年的多模态与Agent,2026年的具身智能——每一次范式切换,都在给上一代人才重新标价。在别的行业,资历是资产;在AI,资历可能是负债。范式两年一换,旧经验变成负资产。

哈特穆特·罗萨在《加速:现代社会中时间结构的改变》里描述过一种“滑坡”:在加速社会里,你必须拼命奔跑,才能留在原地。AI人才,必须跑得比技术更快。

亿邦持续追踪报道该情报,如想了解更多与本文相关信息,请扫码关注作者微信。

本文首发于 亿邦动力 官方网站

文章来源:亿邦动力

广告
微信
朋友圈

FAQ回顾

AI人才的平均司龄为什么只有1到1.5年?

AI公司的组织架构本质上是技术路线的表达,技术路线一变,人才就可能出局;从Transformer、大模型、多模态与Agent到具身智能,每次范式切换都会给上一代人才重新标价。顶尖人才追逐新技术浪潮创业,普通人才则受融资和业务调整影响,因此流动频繁。

AI领域的技术半衰期有多短?

《哈佛商业评论》2023年文章引述哈佛商学院与BCG研究称,职业技能平均半衰期已不到5年,部分技术领域仅2.5年;TechClass数据则显示软件工程与AI专业技能半衰期压缩到18–24个月。模型发版周期从半年缩短到3.5个月,技术方向保质期越来越短。

AI研究员的职业寿命真的比运动员短吗?

比较口径是单一技术方向的有效期。NFL球员平均打3.3年,NBA4.8年,MLB5.6年,NHL约5年,平均退役年龄27到30岁;而AI研究员在单一技术方向上的保质期可能短于NFL球员整个职业生涯。顶尖AI人才平均司龄已降到1到1.5年。

AI行业为什么出现明显的人才年轻化?

AI领域正由二十多岁到三十出头的年轻人掌舵:腾讯姚顺雨27岁任首席AI科学家,混元团队80%是95后;字节曾妍29岁升到4-2;小米罗福莉95后入职10个月升22级;阿里刘大一恒30岁出头接管Qwen。之后他们还要与更年轻、更便宜、更懂新范式的人竞争。

AI创业公司为什么会出现入职1个月到半年就离职的情况?

部分具身智能初创公司没有销售业绩,运转依赖融资,融资不顺就只能裁员;另一种情况是招聘只为达成阶段目标,一旦目标无法实现或事情失去意义,就会裁员或强制转岗。这些员工在职时间可能只有1个月到半年,因此入职AI创业公司像一场前途未卜的豪赌。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0