大魏dawei

Former HR Dir., Jinchuan Group & Telia China｜100% true story 真实记录

探索AI（SI）| 分享经验，讲清边界

[大魏dawei @HRtoAI1]

我是大魏，前外企HR总监，尝试在跨界路上，摸索AI和出海搞钱。关注科技产业。

[大魏dawei @HRtoAI1]

@ameliax111 动手能力很强了

[大魏dawei @HRtoAI1]

@vista8 英文不太好，只能听ai转播

[大魏dawei @HRtoAI1]

@eater188888 也得看情况，大部分是要回一下的

[大魏dawei @HRtoAI1]

@ikun666 这一趟也没白跑，全程实况，给自己也给涨了���经验。

[大魏dawei @HRtoAI1]

@Kling_ai Repeatable AI video production needs more than stable motion. A shared brief, version control and clear approval rights can prevent revision loops from eating the time saved in generation.

[大魏dawei @HRtoAI1]

@NVIDIAAI Factory AI changes jobs as well as machines. Bringing operators into simulation testing early can surface exceptions the model misses and turn deployment into practical on-the-job training.

[大魏dawei @HRtoAI1]

@wallstengine Token growth alone doesn't tell us whether agents are getting more useful. Cost per successfully completed task, including retries and human review, would be a much better business metric.

[大魏dawei @HRtoAI1]

@nvidianewsroom The prepared-site caveat matters. A 14-day deployment still needs power, permits, networking and trained operators ready beforehand. That shifts work earlier rather than making it disappear.

[大魏dawei @HRtoAI1]

@kimmonismus For AI radiology, faster detection is only part of the workflow. Who reviews uncertain cases, how are false positives handled, and does the system reduce the total time to treatment?

[大魏dawei @HRtoAI1]

@GergelyOrosz Hiring can fall into the same trap as architecture: building for a scale the business hasn't reached. Early roles should map to the next customer milestone, not a future org chart.

[大魏dawei @HRtoAI1]

@NVIDIAAI Becoming an AI company also means redesigning work. I'd start with one workflow, name its human owner, and define the quality bar before buying more tools or rewriting job titles.

[大魏dawei @HRtoAI1]

@satyanadella For the NHS rollout, I'd track admin minutes saved per shift alongside error rates and staff workload. Otherwise, time saved in one step can quietly become extra checking in the next.

[大魏dawei @HRtoAI1]

@TTT_News_ @solana The partner-culture point is underrated. Before choosing a platform, I'd ask how incidents are escalated, who owns integration work, and how both teams handle a roadmap change.

[大魏dawei @HRtoAI1]

@nvidia A petaflop is a striking headline. For a small team, the practical questions are which models fit in memory, how fast they run, and what the whole system costs to operate.

[大魏dawei @HRtoAI1]

@Huawei For enterprise buyers, lower latency matters when it changes a workflow. I'd like to see end-to-end task time and cost per completed task alongside the hardware benchmarks.

[大魏dawei @HRtoAI1]

@cgnot996 祝贺

[大魏dawei @HRtoAI1]

@vista8 可以的，不过ai评论也在进步，你这算法还要经常更新

[大魏dawei @HRtoAI1]

@scaling01 OpenAI’s ultimate dream: be the sole AI god, hold all top talent, and make sovereign states bow to tokens pricier than oil.
​Then Sam Altman wakes up in early '25 to DeepSeek-R1—flagship intelligence, 1/10th the inference cost, and fully open-source. Absolute WTF moment. 😂

[大魏dawei @HRtoAI1]

@deanwball OpenAI’s ultimate dream: be the sole AI god, hold all top talent, and make sovereign states bow to tokens pricier than oil.
​Then Sam Altman wakes up in early '25 to DeepSeek-R1—flagship intelligence, 1/10th the inference cost, and fully open-source. Absolute WTF moment. 😂

[大魏dawei @HRtoAI1]

OpenAI’s ultimate dream: be the sole AI god, hold all top talent, and make sovereign states bow to tokens pricier than oil.
​Then Sam Altman wakes up in early '25 to DeepSeek-R1—flagship intelligence, 1/10th the inference cost, and fully open-source. Absolute WTF moment. 😂

[Dean W. Ball @deanwball]

Distillation has always been part of the explanation for why China is so close behind the U.S. in model benchmark performance. But one of my updates after joining openai is that distillation probably plays a bigger role in the explanation than I realized.