On Tue, Oct 6, 2020 at 8:21 AM Matt Mahoney <[email protected]> wrote:
> ... > A contest with prize money needs practical limits on computing resources > so we can test entries. The top entries on my benchmark, even though it > uses the same data, can't be easily verified because it takes 2 weeks of > CPU time and 32 GB of RAM. It does reveal an important fact about AGI, that > our decades of failure to solve it is due at least in part to vastly > underestimating the computing power required. > Network effect monopoly money creates a degenerate environment in which people are rewarded for ignoring algorithmic information theory while they throw vast resources at things like GPT-3 and even boast about how complex their models are as though that's a benchmark when it is an anti-benchmark. This is an example of why a tax on liquidation value should replace taxes on economic activity: ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T6761a13445e5864b-M5259daec9e4f815ab02351dc Delivery options: https://agi.topicbox.com/groups/agi/subscription
