Problem: Siloed Solutions
“Many companies have discovered, often to their
surprise, that it is easy to apply AI and get quick
results. What is not so easy is building a system
of AI applications along with associated data
pipelines that interact and are reliable.”
Source: MIT Sloan Management Review’s 2018 Global
Executive Study and Research Report on AI
Percentage of respondents whose understanding of AI has changed a lot or to a great extent in the past year
A Skills Gap
Big companies are investing heavily in hiring AI talent.
Tapping into platforms and tools for AI “as-a-service” is only as good as the solutions offered by the vendors.
Many Performance Bottlenecks
Serialized code runs like a relay race, one process at a time, limiting the capability and speed of the program. With Parallelized code the work is run like a marathon race, performing work simultaneously creating the most effective and efficient results.
Writing highly parallelized algorithms remains a big challenge for software engineers. The best competing solution achieves 30X compared to our 100X CPU speedups, which we achieve not by writing code, but by generating machine instructions in real-time.
Per Amdahl’s Law:
If the program has 50%, 10% and 1% of serialized code, it is
limited to a 2x, 10X and 100X speedups respectively.
ACCELERATION LIMITS OF PARALLELISM WHEN THERE’S SERIALIZED CODE
“A recent survey we ran at Altimeter found that 38% of consumers rate their understanding of how companies are protecting their privacy as low or extremely low:
45% responded the same of their trust. But, this study also finds that while understanding and trust are low, interest is high—about half of consumers surveyed say they are interested or extremely interested in understanding how companies are using their data.” -Jessica Groopman, Industry Analyst, Altimeter Group