From DALL-E 2 to ChatGPT, masking AI’s wild yr | The AI Beat

It was my first week at EnterpriseBeat, in mid-April. OpenAI had simply launched the brand new iteration of its text-to-image generator, DALL-E 2; our lead AI author, Kyle Wiggers, had moved to TechCrunch earlier than I might decide his mind; and I used to be panicking.

I scrolled frantically by Twitter pictures of avocado chairs and astronauts driving horses on the moon, questioning what all of the fuss was about. I had written about AI tendencies for over a decade, nevertheless it was at a sky-high stage — assume ideas for the C-suite. Now, I belatedly realized how little I understood concerning the previous decade of progress in AI, from machine studying (ML) and laptop imaginative and prescient to pure language processing (NLP). Every beat I’ve ever lined has had a studying curve, in fact. But the AI beat felt like Mount Everest. 

“Give me six months,” I instructed everybody who reached out. “I’ll know a lot more in six months. For now, I really need you to start from the very beginning.” 

A yr of quick and livid AI improvement

Nine months and over 120 tales later, I can look again and see that not solely was I climbing a steep studying curve, however the tempo of reports within the AI area was quicker than in any business I had lined earlier than. 

Who might probably sustain with this wild experience? I did what I might: I lined the largest AI mannequin information, from DALL-E 2 to Google’s Imagen, Meta’s Galactica to ChatGPT. But I utterly dropped the ball on DeepMind’s AlphaFold. I managed to dig deep into DeepMind’s AlphaTensor, with its skill to create quicker novel matrix multiplication algorithms, however was late to masking Stable Diffusion and its quick open-source influence. 

I tackled AI laws and regulation: There have been the AI hiring instruments beneath scrutiny, the brand new AI Bill of Rights and the upcoming EU AI Act. There have been so many tendencies to cowl, from emotion AI and the potential for crippling AI cyberattacks to deepfakes and MLOps. There have been the large model efforts in AI, together with Walmart, Coca-Cola and John Deere, and the large tech information, from Microsoft and Google to Meta and Amazon.

And who might overlook the summer time give attention to AI “sentience”? 

Finally, I had the chance to interview high AI leaders, together with Geoffrey Hinton, Yann LeCun and Fei-Fei Li, concerning the tenth anniversary of the so-called deep studying “revolution.”

It was rather a lot. But it solely made me curious and wanting to do extra in 2023.

Thanks and some humble resolutions

So many colleagues and business leaders have helped me this yr as I bought my sea legs masking the AI beat.

There are far too many to call, however I actually wish to shout-out to EnterpriseBeat managing editor Dan Muse, whom I used to be additionally fortunate sufficient to work with when he was at CIO.com, in addition to EnterpriseBeat founder and CEO Matt Marshall — each gave me the prospect to make the AI beat my very own.

Thank you to everybody I’ve spoken to on the seller facet, within the tutorial neighborhood, on the main consulting companies, on the largest enterprise firms — you’ve all been affected person, supportive and useful.

To my fellow journalists masking AI — together with high writers like Will Douglas Heaven, Melissa Heikkilä, Kate Kaye, Will Knight, Cade Metz, Kevin Roose, Khari Johnson and, in fact, Kyle Wiggers — chances are you’ll not know me, however your protection has taught me a lot. I study from you each day!

As for my New Year’s resolutions, I humbly promise to do my finest to abide by Ben Schneiderman’s pointers for journalists and editors about reporting on robots, AI and computer systems.

I additionally resolve to all the time maintain a newbie’s thoughts, ask the very best questions I can, and, unlike ChatGPT, all the time admit after I simply don’t know.

A contented vacation season and New Year to all!

Feel free to achieve out at sharon.goldman@venturebeat.com or on Twitter: @sharongoldman