Tech Newsround – Meta & DALL-E (AI)

Meta Platforms – double down

• Meta is doubling down on its commitment to the Metaverse by, voicing its intention to allocate 20% of its expenses to the Metaverse in a move that is almost certain to hit earnings hard in 2023.
• Meta’s CTO Andrew (Boz) Bosworth reiterated Meta’s commitment to the Metaverse in a blog post (see here) where he stated that the company was comfortable with this level of spending for the time being.
• I have recently estimated (see here) that Meta’s expenses in 2023 will be around $97.3bn, meaning that spending on the Metaverse is expected to be $19.2bn.
• This is a vast amount of money to spend on a technology that might take off at the end of the decade leading me to think that this is also an insurance policy.
• Apple’s advertising policies have incensed Meta and cemented its determination not to be dependent on the platforms of others which is why it is determined to be a vertical player in the Metaverse.
• This may work out well in the next decade or so but in 2023, this is going to have bad consequences.
• If 2023 revenues are flat at $116bn (they might decline), EBIT will be $18.7bn which after-tax translates into $15bn of net income or $5.58 per share.
• This puts the company on 20.4x 2023 PER which is very expensive for a company where earnings have fallen precipitously with no recovery in sight.
• Furthermore, governance remains poor with Mr Zuckerberg alone in the driving seat meaning that a heavy discount is warranted to fair value when looking for an entry point into this stock.
• Hence, I don’t think the worst is over for this company and with its short to medium-term outlook, $70 per share is where I would begin to get interested.
• There are far better options to look at where there is growth to be had at a much lower multiple

Artificial Intelligence – DALL-E

• After having some fun with ChatGPT (see here), RFM turned its attention to DALL-E to test its ability to draw outside of the box where it utterly failed to perform.
• RFM’s daily newsletter is accompanied by a topical image that is often difficult to match to the theme of the topic under discussion.
• Consequently, DALL-E as advertised is a fantastic use case as there is the potential to create a picture that fits the theme of the day with very little effort on a daily basis.
• However, reality brought this hope crashing to earth as DALL-E in my experience is useless at creating anything that it has not explicitly been taught.
• For a recent newsletter, I wanted a picture of a Tesla racing down the slope of the share price but what DALL-E produced was laughable.
• For another I wanted a picture of a VW Beetle dancing backwards but again, I received gibberish.
• I tried again and asked for a robot playing Risk but got human and robotic hands playing chequers.
• I did manage to get Bitcoins in a freezer but nothing like as good as I can find online with no effort.
• I also asked for a Google robot driving a Renault car through the streets of Paris and received a very early Waymo vehicle with an unintelligible logo roughly superimposed on a blurred Paris street.
• Consequently, I can only conclude that DALL-E for any practical purpose is useless and when put to the test massively underperforms the high expectations that have been set for it.
• I suspect that it has been trained to produce weird and whacky abstract art and when it is asked for something that combines real-world subjects in unexpected situations (as above) it breaks.
• This is yet another example of how these AI models remain unable to deal with generalisation and at their heart are nothing more than pattern recognition and matching systems.
• I continue to think that real hope for progress lies in the combination of software and deep learning which is a field referred to as neurosymbolic AI.
• Meta Platforms made some recent progress in this field (see here) and this is where I am looking for the solution to some of the trickier problems such as autonomous driving.