Learning Library

← Back to Library

IBM AI Enhances US Open, Announces Telm2

Key Points

  • IBM Consulting partnered with the USTA to power the US Open’s digital experience, deploying enterprise‑ready Granite foundation models for large‑scale generative AI content creation.
  • The “content engine” used the Granite‑13B chat model to automatically generate pre‑match bullet points, detailed post‑match reports, and spoken commentary/subtitles by pulling from match statistics and player data.
  • Improvements to AI‑generated audio commentary focused on making the voice sound natural through tuning of top‑K and temperature sampling parameters, as well as careful pitch and speed adjustments, enabling near‑real‑time, largely unsupervised output.
  • IBM announced the next‑generation TELM 2 processor and the upcoming IBM Spire AI accelerator, slated for release in 2025, marking a new era of enterprise computing optimized for AI inferencing.

Full Transcript

# IBM AI Enhances US Open, Announces Telm2 **Source:** [https://www.youtube.com/watch?v=VFEJrddQIb4](https://www.youtube.com/watch?v=VFEJrddQIb4) **Duration:** 00:04:41 ## Summary - IBM Consulting partnered with the USTA to power the US Open’s digital experience, deploying enterprise‑ready Granite foundation models for large‑scale generative AI content creation. - The “content engine” used the Granite‑13B chat model to automatically generate pre‑match bullet points, detailed post‑match reports, and spoken commentary/subtitles by pulling from match statistics and player data. - Improvements to AI‑generated audio commentary focused on making the voice sound natural through tuning of top‑K and temperature sampling parameters, as well as careful pitch and speed adjustments, enabling near‑real‑time, largely unsupervised output. - IBM announced the next‑generation TELM 2 processor and the upcoming IBM Spire AI accelerator, slated for release in 2025, marking a new era of enterprise computing optimized for AI inferencing. ## Sections - [00:00:00](https://www.youtube.com/watch?v=VFEJrddQIb4&t=0s) **IBM AI Powers US Open Coverage** - IBM leveraged its Granite foundation models on Watson X to automatically produce pre‑match bullet points, detailed post‑match reports, and spoken commentary for every US Open singles match. ## Full Transcript
0:00IBM at the US Open and a look at the new 0:02telm 2 processor all on this episode of 0:05IBM Tech 0:10now for more than three decades IBM 0:13Consulting has collaborated with the 0:15United States tennis Association to 0:17provide an engaging digital experience 0:19for US Open tennis fans of course this 0:22year's tournament was no different and 0:23we're going to look at two amazing 0:25generative AI projects that leverage 0:27IBM's versatile family of enterprise 0:30ready Granite Foundation models the 0:32first project was the content engine 0:34which was responsible for providing 0:36up-to-date story coverage for the 0:38hundreds of matches across the men's and 0:40women's singles competition truly a 0:42massive undertaking the content engine 0:45produced three main outputs bullet point 0:47descriptive texts before and after every 0:49singles match multi-paragraph match 0:51reports that provided descriptive 0:53summaries and Analysis about completed 0:55matches and spoken commentary and 0:57subtitles for match highlights the 1:00granite 13B chat model was responsible 1:02for producing our bullet points before 1:05and after each match pre-match bullet 1:07points Drew from Myriad data points and 1:10gave insights based on rankings 1:12head-to-head results and player 1:13biographies when a match was finished 1:16the system generated text descriptions 1:18of what happened drawn from stats such 1:20as Aces break points one double faults 1:22winners and shot speed then the match 1:25reports were created these reports Drew 1:27on trusted Us open data and used the 1:29combined power of granite and other 1:31models hosted on IBM Watson x. to create 1:35long form summaries the second major 1:37project was AI generated spoken 1:40commentary and subtitles for match 1:42highlights introduced last year AI 1:45generated audio commentary provides 1:47automated voiceovers and subtitles for 1:50every singles match highlight reel shown 1:52on the US Open website and app this year 1:55a key goal was to make the audio 1:57commentary more natural and human this 1:59included experimenting with two 2:00variables top K which is a parameter 2:03that controls the number of possible 2:05answers the model should consider and 2:07temperature sampling which was used to 2:09adjust the probability distribution of 2:11possible answers after extensive testing 2:14the next step was going from text to 2:16speech where it was essential to make 2:18the voices sound convincingly human 2:20through experimentation and many 2:22different runs the teams made sure the 2:24voices were clear and had the right proy 2:26in other words the pitch and speed 2:28matched the nature of what was being 2:30said when these balances were found the 2:32inference and output processes happened 2:34largely unsupervised in near real time 2:37to learn more about IBM at the US Open 2:39check out the link in the description of 2:41this 2:42video next up in 2021 IBM introduced the 2:46IBM telm processor featuring its first 2:49Advanced on-processor chip AI 2:51accelerator for inferencing now we're 2:54excited to announce the next generation 2:55of Enterprise Computing for the AI ERA 2:58with the IBM tm2 processor and a preview 3:01of the IBM Spire accelerator both are 3:04expected to be available in 3:062025 developed using Samsung 5 nanometer 3:09technology the new IBM telm 2 processor 3:12will feature eight high performance 3:13cores running at 5.5 gz the processor 3:17will include a 40% increase in onchip 3:20Cache capacity and it will integrate a 3:22new data processing unit specialized for 3:25Io acceleration and the next generation 3:27of onchip AI acceleration 3:30we've also significantly enhanced the AI 3:32accelerator on the telm 2 processor the 3:35compute power of each accelerator is 3:37expected to be improved by 4X reaching 3:4024 trillion operations per second at the 3:44same conference where we announced the 3:45upcoming telm 2 processor IBM also 3:47showcased the IBM Spire accelerator the 3:51Spire accelerator will contain 32 AI 3:53accelerator cores that will share a 3:56similar architecture to the AI 3:57accelerator integrated into the T 2 chip 4:01both IBM telm 2 and the Spire 4:03accelerator are designed to support a 4:05broader larger set of models with what's 4:08called Ensemble AI method use cases 4:10using Ensemble AI leverages the strength 4:13of multiple AI models to improve overall 4:16performance and accuracy of a prediction 4:18as compared to individual models to 4:20learn more hit the link below thanks so 4:23much for joining me today for this 4:24episode of IBM Tech now if you're 4:26interested in learning more about the 4:27topics I've covered make sure you 4:29explore the the links in the description 4:30of this video and of course please don't 4:32forget to subscribe to our channel to 4:34stay up to date on what's going on at 4:36Tech now 4:38[Music]