Lead story - Generative AI in the enterprise - project myths and realities
I put a bit of a stake in the ground with Enterprises will test the limits of LLMs - and ChatGPT is just the beginning. It's a blog post, not any kind of quadrant, but it's also a mile marker in what I've learned in the generative AI research hustle so far. It's a statement of where I believe the shortfalls - and use cases - lie:
Enterprise AI vendors are determined to raise the bar on the flawed performance of ChatGPT-type output, where models are trained on the entire Internet - and imperfect guardrails then imposed. But will these vendors succeed? If so, which ones?
To help find out, I put a slew of vendors to the AI PR challenge. A few stepped up with some useful/different positions. Questions we are pressing this fall include:
- How are enterprise LLMs being trained?
- How are beta customers faring? What use cases have early traction?
- Will early customers will pay a premium for AI features? How will that premium go over, if those same customers are investing their own data - and co-innovation time - training enterprise models?
- How comprehensive is the approach to reducing model/data bias?
Then there is the million/billion dollar question:
With the help of enterprise LLMs and customer-specific refinements, I believe we can move beyond "instant mediocrity" use cases. But human-in-the-hoop design can't take the rough edges off of every use case. Some generative AI use scenarios truly have an outlier problem - where the inevitable outliers have too much downside to pursue.
I pressed these issues on my just-released podcast: Generative AI in the enterprise - project myths and realities with Vijay Vijayasankar. One keeper quote from Vijayasankar, on the virtues of narrow/industry specific models:
I'm more or less firmly convinced, based on the last several months of client interactions and knowing what our own engineers are working on, that narrow models are far superior in achieving goals for enterprises. Because most of the time enterprises don't need the consumer type applications, right? There are no scenarios where you need to answer a question on FDIC regulation, and then have the same model create an essay or write about your cat. So you might as well focus on finding the business case where you have the most revenue generation - or most cost saving - or whatever business case [is right for] your company, and use very specific solutions.
How well those "narrow" LLM solutions pan out for customers will be one of the biggest AI questions/potentials of 2024. The podcast also gets into whether bias and hallucinations can ever be eliminated from LLMs, as well as some of the most promising use cases Vijayasankar is seeing in his financial services purview - and why.
On diginomica, Gary also advanced the generative AI conversation, via Dun & Bradstreet - accurate data must be the basis for any serious enterprise use of generative AI. More industry use cases are cited, including a procurement-focused, first-level support bot. As Gary notes, explainability remains a crucial goal/requirement. He quotes Dun & Bradstreet's Chief Data and Analytics Officer:
You need to be sure that you can always connect back to your data to understand where any answer came from.
In other words, you not only need to see the answer - you are also going to want to know the underlying document or publicly available source or any other private source that produced the answer your AI’s giving you.
That sets the issues for the next round of enterprise events, as the diginomica team fans out again. I'll be at Workday Rising San Francisco next, pesky AI questions in hand.
Vendor analysis, diginomica style. Here's my three top choices from our vendor coverage:
- Cisco spunks $28 billion on Splunk - CEO Chuck Robbins explains why - It's Cisco's biggest acquisition ever, so 'why' is a fair question. Stuart quotes Cisco CEO Chuck Robbins: "The combination of Cisco and Splunk further enhances and accelerates our strategy to securely connect everything to make anything possible. The IT landscape is changing faster than we've ever seen, with hyper-connectivity, AI and increasing cyber threats. The value of data only increases and that's why this deal makes sense." Cisco's share price fell 4 percent after the deal, indicating that this is no small gambit - no matter the buzz factor behind data platforms and AI.
- Workplace culture - O.C. Tanner & new Management Thinking - Brian reports back on his forays at O.C. Tanner's user event: "O.C. Tanner is a major player in the rewards and recognition space but it also has research capabilities and other assets that help their customers improve their corporate culture, employee engagement levels, employee experiences, etc. They are more than a swag shop or a producer of 'atta-boy/girl' certificates."
- 'Who doesn't want to know what's happening in real time?' - Tableau CEO Ryan Aytay on democratizing analytics - Phil on his sit down with Tableau's (relatively) new CEO: "The new tools change the process from one in which the data specialist goes away and designs a report or dashboard to order, to become one in which the role of the data specialist is to design ready-made templates and tooling that business users can immediately pick up and use, while standing by to offer assistance when help is needed to plug in new data sources or fine-tune the results. It's much more of a continuous collaboration between specialists and users."
Oracle CloudWorld 2023 coverage and analysis - last year, Larry Ellison's incredibly ambitious global healthcare goals were the (heavily debated) CloudWorld headline. This year, Ellison refined his healthcare vision, but, no surprise, AI was the central theme. Derek and I were on the ground, pressing the questions and documenting customer use cases. Phil and Stuart contributed from the virtual event stream, and there is more analysis to follow (I also did a CloudWorld 2023 podcast review with diginomica contributor Brian Sommer). For now, here's a few picks from diginomica's CloudWorld coverage:
- CloudWorld 23 - Larry Ellison’s Oracle AI vision extends well beyond the enterprise, it’s a vision for big world problems - Derek
- CloudWorld 23 - How Native American Tribe Twenty-Nine Palms migrated all its systems to Oracle to support future growth - Derek
- CloudWorld 23 - Move over Darling...to standardized global cloud HCM - Stuart
- CloudWorld 23 - Steve Miranda on why the financial supply chain matters to healthcare, and what's next for generative AI - Jon
Jon's grab bag - A political/tech storm surged in the UK via a controversial new Net Zero policy. Stuart weighed in with No, Prime Minister - your chaotic Net Zero political games are bad news for UK tech sector investment. Em also had choice words in My take - why the UK’s Net Zero delay is incompatible with ambitions for tech industry, AI leadership…and life on earth.
Neil added a geopolitical angle in Chinese whispers - what are the enterprise ramifications of Biden's Executive Order to restrict investment in sensitive technologies in China? Meanwhile, Madeline continued her insightful series with What I’d say to me back then - tech entrepreneur Heather Shoemaker on overcoming sexism in the VC market.
Best of the enterprise web
My top seven
- Cisco to acquire Splunk in $28B mega deal - Ron Miller reported on the initial news; Andy Thurai brought context and raised lingering questions in HOT TAKE: Cisco is buying Splunk for $28 B in cash. What does it mean for you as a customer?
- Why Cradlepoint's acquisition of Ericom predicts the future of SASE in the enterprise - Louis Columbus bears down on the security beat - and for those who love acronyms, SASE = Secure Access Service Edge, or, if you prefer, converged networking and security as a service isn't far off.
- Is there really an information security jobs crisis? - Anyone who pokes holes in the so-called "jobs crisis" narrative is someone I want to hear more from. Ben Rothke takes it one step further, explaining why certain security jobs are tougher to contend with: "Kusher notes that there is not a shortage of security generalists, middle managers, and people who claim to be competent CISOs. Nor is there a shortage of thought leaders, advisors, or self-proclaimed cyber subject matter experts. What there is a shortage of are computer scientists, developers, engineers, and information security professionals who can code, understand technical security architecture, product security and application security specialists, analysts with threat hunting and incident response skills. And this is nothing that can be fixed by a newbie taking a six-month information security boot camp."
- Reflection: Advice for Supply Chain Technology Leaders Making Today’s Technology Pivot - Lora Cecere's latest includes sharply-worded views on where generative AI does - and doesn't - fit: "Step 7. Use Generative AI for Content Development. Shipping instructions. User guides. Training materials. These are all good use cases for generative AI. I don’t think we are ready to use generative AI as an engine in supply chain planning. The reason? The organization is not aligned on the definition of supply chain excellence. Alignment issues grew three-fold over the last decade with a rise in politics in large organizations."
- Why open source is the cradle of artificial intelligence - is AI the province of big tech, or is open source going to be a major factor? Steven Vaughan-Nichols makes the open source case.
- The Technology Facebook and Google Didn’t Dare Release (registration wall, NY Times) - a disconcerting deep dive on tech the world may not be ready for, or even want: "How did we get to this point where someone can spot a 'hot dad' on a Manhattan sidewalk and then use PimEyes to try to find out who he is and where he works?"
Good news/bad news: if you're headed for an event, expect a full/fun flight. Good news: you can bring your light saber:
Light Saber | Transportation Security Administration https://t.co/L9K8utCbTG
-> good to know that the TSA has a sound policy on lightsabers ) cc: @brianssommer
— Jon Reed (@jonerp) September 25, 2023
This might be a bit too obvious for a full-on whiff, but Frank Scavo kicked tires on ChatGPT's music service and found that Elvis Presley is (way) too high a bar:
Checking out samples of the music generation service of OpenAI, called Jukebox. It's pretty bad. The music is bad, but the lyrics really make no sense. If you don't have much time, check out the 2nd one "Rock, in the style of Elvis Presley" @jonERP https://t.co/sVnr9R32Hf
— Frank S. Scavo (@fscavo) September 24, 2023
As I replied:
can't even be bothered by it to be honest, genai has cool use cases and this most definitely isn't one of them. meanwhile all kinds of amazing humans are doing stuff like this - https://t.co/yltGDWQu7r
— Jon Reed (@jonerp) September 24, 2023
Since I'm out of whiffs, how about a scorching find from YouTube's scary-addictive recommendation album instead:
Orianthi - Voodoo Child https://t.co/xFeZ13MR7m
-> god bless the YouTube algo, wow!
— Jon Reed (@jonerp) September 13, 2023
See you next time... If you find an #ensw piece that qualifies for hits and misses - in a good or bad way - let me know in the comments as Clive (almost) always does. Most Enterprise hits and misses articles are selected from my curated @jonerpnewsfeed.