AI 目前面临的困境主要包括以下几个方面：
企业的重点主要放在自主构建应用程序上，目前企业更倾向于自主开发应用程序，而不是从外部购买。这主要是因为市场上缺乏经过严格测试、能够在特定领域内取得决定性成功的企业级人工智能应用程序。毕竟，目前还没有像这样的应用程序的“魔法象限”（Magic Quadrants）。基础模型的出现使企业更容易通过API来构建自己的AI应用程序。企业现在正在构建自己版本的常见应用，如客户支持和内部聊天机器人，同时还在尝试更新颖的英语，如编写消费品配方、缩小分子发现范围和进行销售推荐。关于“GPT wrappers（GPT套壳）”的局限性已经被讨论过很多了，例如，初创公司利用大型语言模型（LLM）已知能力（例如文档摘要）来构建用户熟悉界面（例如聊天机器人），我们认为这些公司将面临的一个困境是，AI进一步降低了企业内部（in-house）自主构建类似应用的门槛。
The scariest thing about cognitive atrophy is that it’s hard to notice in the moment.With physical muscles,I can feel a difference.A year ago,I got busy and stopped lifting weights for two months.I felt a difference when I lifted a couch or when I was carrying a lot of groceries.I knew,in my daily life,that I needed to get back to the gym.But in the case of Google Maps,it happened without me realizing it at all.I thought I had retained my spatial reasoning skills because I had used them for two decades.But then,they evaporated and I didn’t noticed it until it was too late.As educators,we want to take a vintage innovation approach that embraces the overlap of old school tools and new technology.We want to embrace the overlap of best practices and next practices:But this requires an intentionality that can be a challenge in a tech-infused world.We start using auto-fill in Google and don’t realize we’re using it.We go to ChatGPT to design some scaffolds and supports and we don’t realize that we have stopped thinking intentionally about it.One potential solution is to track how often we use AI.As educators,we can do a time audit where we track how often we use AI versus engaging in a fully human-centered approach.We can step away from the tech and ask if we are growing too dependent on machine learning.We might even choose deliberate times to be fully tech-free.In the end,cognitive atrophy will be one of the most significant challenges of AI.As educators,we need to be intentional about how we use it professionally and with our students so that we don’t off-load the thinking to a machine.We need to draw students into these conversations as well so that they can learn to use AI wisely as they navigate an uncertain future.
成为某个领域顶尖人才通常以多年的密集信息输入开始，通常是通过正规的学校教育，然后是某种形式的学徒实践；数年时间都致力于从该领域最出色的实践者那里学习，大多数情况下是面对面地学习。这是一个几乎不可替代的过程：例如，医学住院医生通过聆听和观察高水平的外科医生所获取的大部分信息，是任何教科书中都没有明确写出来的。通过学校教育和经验，获得有助于在复杂情况下确定最佳答案的直觉特别具有挑战性。这一点对于人工智能和人类都是如此，但对于AI来说，这个问题因其当前的学习方式以及技术人员当前对待这个机会和挑战的方式而变得更加严重。通过研究成千上万个标记过的数据点（“正确”和“错误”的例子）——当前的先进神经网络架构能够弄清楚什么使一个选择比另一个选择更好。我们应该通过使用彼此堆叠的模型来训练AI，而不是仅仅依靠大量的数据，并期望一个生成模型解决所有问题。例如，我们首先应该训练生物学的模型，然后是化学的模型，在这些基础上添加特定于医疗保健或药物设计的数据点。