🔮Memia 2024 Predictions Scorecard
A decent enough 75% pass mark to start 2025 off with
Kia ora from Aotearoa New Zealand to all Memia readers,
Wishing you all the best for a productive, prosperous (in every sense of the word) and above all *peaceful* 2025. It’s certainly going to be a wild ride.
As is my rhythm for this time of year, I’m enjoying my southern hemisphere summer holiday with family and friends, once again camping in the stunning Marahau valley at the south end of Tōtaranui/Abel Tasman National Park. Here is the view of the tractors and trailers lining up to pick up the returning water-taxi boats at low tide this morning:
The first weekly newsletter of the year is currently in production, but first I thought I’d send this post out … how did I do with my predictions from last year??
🔮Memia 2024 tech and environmental predictions, scored
Way back in January 2024 in Memia 2024.01 — a whole year ago🤯 — I brain-dumped my 59 Tech and Environment Predictions for 2024. In this post I look back with a critical eye and mark my work…
(Note: I pasted the finished post below into Claude (Sonnet latest) and ChatGPT (o1-mini) and asked them to count up the totals…they both gave me different answers, so I had to manually go through my working… and they were both wrong, so much for AGI 🤣)
✅ Correct Predictions 32 (54.24%)
🔶 Partially Correct 12 (20.34%)
❌ Incorrect Predictions 15 (25.42%)
…Which gives me a pass mark of 74.6% correct or partially correct. I’ll take it 😇, particularly as some of the incorrect predictions were clearly marked “speculative” at the time… must remember not to do that next time! :-)
Scorecard summarised:
(Check the exact wording of the prediction in the text below for details…)
Detailed commentary follows: predictions pasted verbatim from last year in italics, scores / comments in bold). Of course, there were a whole bunch of things that I missed completely as well… for discussion in another post, perhaps....
2024 Tech Predictions
🧠AI
The good
The accelerated pace of Generative AI innovation from 2023 will continue. Expect weekly releases of all sorts of new GenAI modes and tools which enable creators and knowledge workers to operate across any possible field of information:
Large “multimodal” models (LMMs) will mostly supercede Large Language Models (LLMs)
✅ (OpenAI’s GPT-4o (omni) sets the bar for multimodal performance — although everyone still calls them LLMs)
Leading commercial labs OpenAI and Google will continue to lead the field with their GPT-series and Gemini models.
✅ (Anthropic was the surprise challenger throughout the year, with Google pulling Gemini 2.0 out of a hat right at the end of the year)
Apple will enter the fray, offering LMM functionality on their own devices (effectively “Siri 2.0”) with added Apple privacy secret sauce and powered by Apple’s proprietary silicon.
✅ Spot on. Apple Intelligence launched by the end of the year with Apple “Private Cloud Compute” promising to perform encryption and performance contortions.
Specialist image and video generation labs including Midjourney, Runway and Pika will stay out front delivering mind-blowing capabilities for cents on the dollar.
🔶 Mostly correct, although I didn’t anticipate a raft of new frontier labs appearing including open source challengers Black Forest Labs (FLUX) and challengers KLING and Genmo. By the end of the year, OpenAI Sora / Google Veo were up there setting the pace for generations and workflow tools.
The gap between commercial and open source models will close even further, led by firms including Meta, Stability AI, Mistral and Hugging Face. By the end of the year a “GPT-4 equivalent” (pick your benchmark) open source model will be easily downloadable and will run locally on a high-end Macbook. OpenAI may even release its legacy models under some variant of open source licence.
✅ Meta’s open source Llama 3.3 gives both GPT-4 and GPT-4o a good run for their money on benchmarks … and using open source decentralised AI frameworks such as Exo Labs you can run even the 405B models on one or more consumer devices ….although you may struggle to run either of them locally with adequate performance. And then right at the end of the year Deepseek 3.0 blew everyone’s minds with its capabilities matching up to OpenAI’s o1 “reasoning” model.
AI agents will become far more powerful and gradually move into the mainstream, able to complete higher and higher abstraction tasks and long-running workflows to human-equivalent levels. Expect to be able to ask ChatGPT, Google Assistant (Gemini), Perplexity or Apple to completely manage the booking of your next family holiday by the end of 2024.
❌ AI Agent frameworks were being released thick and fast towards the end of the year and so far specialist AI agents for software development (eg Devin) were starting to appear - but a generalist AI agent is still yet to arrive…. but soon.
GenAI and AI agents will become embedded into the productivity software which we use every day… anyone not using these tools will increasingly find their role challenged by super-productive human+AI “centaurs”. Expect whole classes of knowledge workers to just be let go in 2024. (Latest example: language learning company DuoLingo “off-boarded” many of its translation contractors in December - the remaining workers will just QA AI-generated translations).
✅ Show me a software product that doesn’t have an AI chat interface now?! (And for software development the assistant functionality integrated into the IDE — Cursor leading the charge — makes productivity shoot up for those using it.)
The first full-length AI-generated movies will be released with budgets of only a few thousand dollars. One or two will be really good.
❌ Again, probably just a few months too early in my timing estimate. 2024 did see completely AI-generated adverts and short films… but not yet a feature-length movie.
The Humane AI pin will disappear into the “too early” history pile. There will be several other attempts at creating an “AI-native” device form factor. None will quite catch.
✅ The early-2024 race to create AI-native devices (Humane, Rabbit, OpenInterpreter) failed to live up to hype and faded after initial enthusiasm.
Note that former Apple iPhone designer Jony Ive has confirmed he is working with OpenAI on a future AI device… but no details so far...
Most interesting: 🐘The Ark - this startup is building a ruggedised, off-grid, solar powered device designed for after the apocalypse… the device will contain maps of the entire world, GPS, fine-tuned 8b AI models for medical questions and rebuilding civilisation and a LoRa communication module.
“AI companions” will become common in retirement homes to alleviate loneliness and boredom in older generations.
✅ Lonely seniors are turning to AI for friendship: Why Nana’s new BFF may be a robot.
Domain-specific generative AI models across science, engineering and design (eg. physics, chemistry, biology, nanotech, health, meteorology, astronomy, structural engineering, architectural design …. and thousands of other subject areas) will accelerate scientific discovery and applications beyond anything previously imagined in 2023. Bring it on.
🔶 Again, we saw big steps on this road but perhaps my projected timing was a year too early on the impacts…
The expensive
A venture funding crunch will drive consolidation among frontier AI labs and the usual Big Tech giants will start acquiring some of the smaller companies.
🔶 Venture funding showed no signs of crunching… but most AI investment went to Nvidia, infrastructure buildout and a handful of leading labs (OpenAI, Anthropic, Google, Meta and XAI)
✅ Smaller frontier labs started to fold… RIP Stability AI (now under new management), Inflection AI (“acqui-hired” by Microsoft) and Adept (“acqui-hired”, by Amazon). The acqui-hires were gymnastics to avoid anti-trust regulation.
Google will somehow manage to maintain its “front page to the web” search engine monopoly, integrating Gemini search capabilities into its product to head off challengers like Perplexity and, er, Bing.
✅ At the end of the year Google’s search engine market share was still up around 90% despite strong challenges from Perplexity and OpenAI. Google’s own AI Answers caused the giant grief mid-year but have become more stable
However Google’s actual search engine usability will decline (when half the links above the fold are either AI-generated spam trying to game the ranking algorithm or paid for).
✅“AI Slop” is causing big challenges for all internet content indexing, including Google
As a result, Google’s huge profit margins from its search advertising unit will lower in 2024.
❌Completely wrong - Alphabet (Google’s parent)’s operating margin for Q4 2024 was 27.5%, up from 24.5% in Q4 2023 … and most of that came from search advertising.
Generative AI copyright infringement cases will go through US courts … and subsequently be appealed and re-appealed… lots of work for lawyers. Yes, copyright holders may win some early skirmishes… but in the medium-long term AI labs will continue developing complex technical workarounds to obfuscate applicability of any resulting case law. And the open source AI community will rapidly evolve its own conventions/workarounds driven by the need to keep up with commercial AI labs. Fundamentally: software development moves exponentially faster than the legal system.
✅ Although approximately 25 lawsuits related to AI and copyright were progressing through various US courts alone at the end of 2024 - the major labs have been testing all sorts of legal arguments to frame their use of copyrighted materials as transformative fair use. Meanwhile massive open source training sets abound and major labs are claiming to have “run out of data” and hence need synthetic data to train the next generation of models. The only people winning are the lawyers while the technologists just largely develop through and over the arcane, archaic concept of “copy” “right” for just a few tokens in 15 trillion…
The costs of using AI will go up: scaling GenAI inference out to the rest of the world will increase energy demand hugely - AI could soon need as much electricity as an entire country. These costs will be need to be passed on to end-users. (2024 will also see big announcements in improved energy consumption and sustainability of chips used for AI training and inference…)
✅ The biggest story of the year is the massive capital buildout required to fuel future AI…
✅ Nvidia, Cerebras and AMD all announced new chip generations with significant energy efficiency and performance improvements (Cerebras claiming 20X Nvidia H100s in November!)
❌ However the actual marginal cost of intelligence continued to fall exponentially during the year as performance improvements and “small” models keep delivering lower-power, lower-compute solutions. And open source labs keeping the leading labs honest.
✅ That said, OpenAI’s price point of US$200/month for its ChatGPT Pro tier (with unlimited access to its o1 and o3 reasoning models) perhaps shows the way for 2025…
This will initiate a drive towards smaller AI models which are more energy-efficient - “small” language models such as Microsoft’s Phi-2 will be utilised in specialist subject domains and provide the same performance as LLMs.
✅ As above - Claude Haiku, Gemini Flash, OpenAI o1-Mini and a raft of 1B- and 8B- parameter open source models continued to improve the price/performance over the year
Already you can run LLaMa (and derivatives), and other open source models locally (disconnected from the internet) on your smartphone: by the end of 2024, a whole spectrum of AI models will be available to download and run locally, disconnected. Effectively: the “compression of all humanity” on your own phone or laptop.
✅ As above … and in fact the Ark, on a little ruggedized handheld device
“AI app stores” such as OpenAI’s GPT-store will proliferate. But sustainable revenue models from generic GenAI applications will remain elusive — so far the only firms making any money from GenAI are chip makers (NVidia) and hyperscale cloud providers (Microsoft/OpenAI), this conundrum could well continue.
🔶 Actually the AI app store model didn’t really catch on in 2024… but the fundamental question about how to make more revenue from AI than you’re spending if you’re not Nvidia remains unanswered!!
The ugly
GenAI bank scams will scale up: millions of people worldwide will be caught and persuaded to send money urgently to their “relative” calling in an “emergency”.
🔶There have certainly been increasing incidents of deepfake bank scams, but not at the scale anticipated. Biggest haul was when scammers posed as British firm Arup's CFO and other employees using deepfake identities on a videoconference call and tricked an employee into transferring US$25 million to Hong Kong bank accounts.
GenAI use in political campaigns will scale up, leading to ever-more effective epistemic attacks on consensus reality, with increasing numbers of people relying on persuasive AI models for their news and political beliefs. (More than 60 countries representing half the world’s population will go to the polls in 2024, the biggest election year ever.)
🔶Again the use of AI in political campaigns did happen, but not at the scale anticipated. The influence of X’s Musk- and Trump- boosting algorithms on swinging the US elections, however, showed the vulnerability for democracy of what are euphemistically termed “low information voters”.
The flipside of the “AI companions” mentioned above: “AI Influencers”, “AI counsellors”, “AI girlfriends”… all will be unleashed on the forming minds of younger (pre-teen, teen and 20-something) generations… with no idea of the longer term consequences.
✅ One suicide was linked to AI — after a Florida 15-year-old formed an emotional attachment to an AI chatbot named "Dany," inspired by a "Game of Thrones" character. Former Google CEO Eric Schmidt expressed concerns that AI-powered "perfect girlfriends" could exacerbate loneliness among young men
One corollary: expect fertility rates to plummet even faster than predicted across the whole planet.
✅ Here’s looking at you, South Korea
The first recorded human death from an autonomous (AI-operated) weapon will be confirmed, which will open the floodgates of autonomous drone swarms and completely transform wargaming into combat trials between remote-controlled intelligent robots. (But will grinding real-world territory-based human wars such as Ukraine/Russia be resolved as a result? Not sure…)
❌As of December 2024, there are no confirmed cases of *fully autonomous* AI weapons being used to kill in warfare. However, there have been plenty of reports of AI-assisted targeting systems and semi-autonomous weapons being deployed in recent conflicts, in particular:
In Gaza, the Israel Defense Forces (IDF) reportedly used an AI-enabled data system called Lavender to select human targets based on behavioral patterns with minimal human oversight.
In Ukraine, autonomous drones have been used to target individuals, though a human still directs them to engage.
The speculative
Strongly persuasive AI systems could lead a group of people en masse into extremist beliefs and actions which result in fatal consequences. (cf. Neal Stephenson’s “epistemic ground zero” Moab event in “Ameristan” in his novel Fall, or Dodge In Hell).
❌ (So far… arguably the entire 2024 US election could be framed this way…)
Finally: OpenAI may finally release multimodal “GPT-5” (or another moniker) and simultaneously announce they have “achieved AGI” (again, pick your benchmark) but by then the concept of “AGI” will be so watered-down that everyone will just look up briefly and carry on.
🔶 I’m claiming 50% on this one. o3 released by the end of the year demonstrated extreme super-human performance on so many cognitive tasks that the term “AGI” increasingly becomes a meaningless term... dissolving into a collection of scores across a multitude of “benchmarks” of various aspects of intelligence. For all intents and purposes, AI is pretty much at or above the economic level.
(See also OpenAI’s interesting definition of AGI in their agreement with Microsoft…)
🕶️XR
The obvious
Apple’s premium US$3500 Vision Pro device will launch on February 2. Market reception will be “hmmmm” - lots of long articles about why XR won’t work due to the psychology of wearing something on your face.
✅ Yep. And US$3500 is clearly too much!
However Meta’s Ray Ban smart glasses, quietly launched at the end of last year, seem to be proving the category for head-mounted cameras that look like specs.
✅ Also yep.
At the other end of the market, Meta will launch their next Quest headset, code-named Ventura, rumoured to be in the US$200 price range.
✅ Hello, Quest 3S for just US$299
The uncertain
If Meta gets ahead of its roadmap, watch out for the first display-enabled smart glasses, which will include a wristband to control the headset, as 2024’s must-have Xmas gift.
🔶 Meta’s “Orion” glasses and wristband were indeed showcased by Zuck this year…but don’t expect it in your Xmas stocking until 2028 at the earliest…
Meta’s ultra-realistic VR avatars will get their first open release. The ability to collaborate in VR (“Zoom+”) will become common by the end of the year and Meta will announce support for Microsoft Teams VR conferencing for its Quest 3 headsets.
❌ Meta continues to push out improvements to its avatars in its Horizon VR worlds and “Workrooms” … but adoption so far pretty low and the ultra-realistic Codec avatars aren’t out of R&D yet.
Also Apple’s Persona avatars which shipped with the Vision Pro are too “uncanny valley” to be usable … yet.
XR headset sales will start to eat into sales of big-screen TVs. It will simply be better experience to watch movies on the “big screen” inside your headset.
✅ Overall global TV sales were forecast to experience a slight decline in 2024, with total 195 million TV units shipped (however, sales of XXL TVs exploded). Meanwhile global shipments of VR and mixed reality (MR) headsets reached approximately 9.6 million units in 2024 (a YoY increase of 8.8%) … so I’ll chalk this up as a pass.
The speculative
A challenger firm will come out of stealth with a paradigm-shifting XR productivity experience. (My eyes on: Unison). Seamless mixed reality combined with AI will augment productivity.
❌ Absolutely no sign of anything like this… Unison still in stealth.
At some point, being in eXtended Reality will be an order of magnitude richer, *more* engaging experience than reality. Not in 2024, but soon. And at that point, why would anyone ever come out again?
*Not* in 2024, I said…. 😆
👥Social media
Twitter / X will keep going and will look completely different within 1 year, with Musk’s promised “everything app” features being released throughout 2024. But it won’t grow market share significantly.
❌ Woah boy… how X has changed in 2024, but not in the UX, which remains remarkably similar to the legacy Twitter. The “everything app” features are still MIA… but the feed algorithm is unreliably compelling once you judiciously mute certain accounts and words… and the addition of Grok into the UX, while still underwhelming, could yield really interesting tools for parsing rapidly unfolding events and trends in 2024.
🔶Meanwhile X's monthly active user base actually increased in 2024 from 556 million in 2023 to approximately 611-619 million, estimated market share only grew 11.37%-11.9% … (but how many of them are bots…?)
Fully decentralised social media will come closer - Meta will rollout Threads support for the Fediverse (eg. implementing the W3C-governed ActivityPub protocol). Mastodon is already there - but Bluesky, Farcaster, Substack, and others will follow suit and converge on ActivityPub support as well. (X won’t).
✅ Meta did indeed start rolling out Fediverse support on Threads. But no-one really uses the app?
Establishing truth / veracity / identity on social media platforms will continue to be a technology and governance challenge, balancing centralised (=censorable) vs decentralised (=permissionless, uncensorable) approaches across a wide spectrum.
✅ Don’t believe anything you read online. Regulators are attempting to come to grips with the challenge… Vietnam mandating user identification on Facebook and TikTok… while Australia is contemplating an all-out social media ban for under-16s. (Good luck with that…)
Meta will continue to make a motherload of cash from its social media advertising platforms.
✅ Meta's revenue showed consistent growth throughout 2024, with Q3 2024 revenue of US$40.6 billion representing a 19% YoY increase. Good job, as this enables them to sink $billions into training their open source Llama models…
🏋️Longevity
The obvious
“Rejuvenation athlete” Bryan Johnson will launch a low-cost “Blueprint” subscription food brand in the US which provides his open-sourced broccoli-and-olive-oil diet to the masses for less than the typical weekly grocery bill.
✅ Yep. Including “snake oil” EVOO …. LOL.
AI will be used to build large models of major diseases including cancer and atherosclerosis (heart disease) which will yield a few major breakthroughs in disease prediction and treatment.
✅ Yes - for example Harvard Medical School researchers created a versatile AI model called CHIEF that can perform multiple diagnostic tasks across 19 cancer types, including detecting cancer cells with 94% accuracy and forecasting patient survival.
A raft of new “personalised” gene therapy treatments will enter trials to provide interventions against the most common genetically-inherited diseases.
🔶As of Q3 2024, 32 gene therapies (including genetically modified cell therapies), 34 RNA therapies and 68 non-genetically modified cell therapies had been approved by the US FDA, with nearly 5000 in development. But progress on FDA approval was slow…
“Longevity Bro” will become a social movement - expect to see lots of 40-50 year olds cutting out alcohol, calorie restricting, HIITing and knocking back a pharmacy-load of supplements every morning. I’m there.🤣
✅I think Bryan Johnson knocked it out of the park in 2024 with his DON’T DIE branding and reach… expect more of this in 2025.
The speculative
The aforementioned Bryan Johnson will announce that he has actually slowed his rate of biological aging across a range of measures to zero. He may even reverse it.
❌ Nope. But 0.61% is still a pretty impressive achievement, assuming that it’s a scientifically meaningful measurement he’s claiming.
🔬Materials science
The speculative
After the hype of last year’s scientific preprint from a team of Korean science team claiming to discover a new compound “LK-99” which functions as a superconductor at room-temperature …but which was not replicable — in 2024 such a compound will be discovered (at least one which functions far closer to “room temperature” than all existing superconductors). An AI model will be involved in the discovery.
❌ In my defence, I was led astray by
😆
2024 Environmental Predictions
🥵Climate [tech]
OK, time to put on my doomer hat…
The scary
To be honest, I’m scared. 2023 saw a marked acceleration in global temperature rise and if 2024 does the same then it’s going to be clear that the IPCC AR6 consensus models are way too conservative. I expect we’ll see some independent science demonstrating this.
✅ One study by EPFL scientists found that climate models predicting catastrophic stronger heating than the IPCC's most probable estimate are plausible and should be taken seriously.
Despite aspirations to the contrary and a rapidly-shifting price difference between fossil fuels and renewable electricity generation, global CO2 emissions will increase for one more year (before finally showing a small decline in 2025).
✅By November, the 2024 Global Carbon Budget project was forecasting fossil carbon dioxide (CO2) emissions of 37.4 billion tonnes, up 0.8% from 2023, with some significant regional variations:
China (32% of global emissions): 0.2% increase
US (13% of global emissions): 0.6% decrease
India (8% of global emissions): 4.6% increase
European Union (7% of global emissions): 3.8% decrease
Rest of the world (38% of global emissions): 1.1% increase
The prevalence of extreme weather events will continue in 2024. Another year of floods, droughts, wildfires, landslides, snowstorms, hurricanes, typhoons. The El Niño weather pattern will continue until at least half way through the year and, related, the Panama Canal will need to close for large periods due to drought, with knock-on effects for global supply chains.
🔶 The Panama Canal didn’t close completely, operating instead on restricted shipping volumes from January to July. But the first annual report by NGO World Weather Attribution recorded climate change caused 41 extra days of dangerous heat in 2024, threatening health globally, with extreme weather killing 3,700+ people and displaced millions across 26 studied events.
The speculative
Last year I predicted that the first state-backed geoengineering initiatives will be announced, which to my knowledge didn’t happen.
❌(Once again) I’ll bring this one forward to
20242025. (The first outdoor test of cloud brightening technology in the US was conducted to explore its potential for reflecting sunlight back into space)
Very speculative, but not beyond the realms of techno-optimism: new domain-specific AI chemistry models will yield a previously unknown, *safe* chemical pathway which is able to use sunlight to effect “artificial photosynthesis” and “grow” useful, economically valuable materials by extracting CO2 (and methane?) directly from atmosphere. And this process will be open-sourced, so that a new economy of “growing” useful construction materials and/or foods at scale will yield a sudden drop in global CO2 concentrations.
❌ No breakthrough “artificial photosynthesis” chemical pathways that I’m aware of in 2024
(But of note: Terraform Industries’ "Terraformer" system combines CO2 capture, hydrogen production, and fuel synthesis, with the company claiming their technology can absorb CO2 for less than US$250 per ton, significantly lower than other methods. The entire system is designed to run on intermittent solar power and produce carbon-neutral methane at competitive prices.)
🌳Biodiversity loss
The ugly
Net global deforestation will continue to track off-target for 2030 net zero forest loss, increasing again in 2024. This will be monitored by new high-precision satellite data which can also track wildfires, pest outbreaks or illegal logging activities.
(Tentative)✅ The data isn’t in for 2024… but these charts from the new WRI deforestation tracker for 2023 tell the story…
WRI Deforestation and Forest Restoration Target Paths (2023 Data)
Global species extinctions will accelerate again in 2024. What can we do here in time?
🔶 Not enough data at this time.
But the 2024 Living Planet Index 2024 reported a 73% average decline in wildlife populations in the 30 years from 1970 to 2020. But with some subtleties inside those numbers:
The speculative
At the same time as net deforestation goes up, new grassroots “Regeneration Ark” initiatives worldwide begin to turn land previously used for human activities back to nature, aiming towards a global goal to conserve at least 30% of the planet.
✅Not enough data in yet to quantify progress in 2024, but the Global Rewilding Alliance is just one of the organisations fighting against the prevailing currents:
Scientists will announce that they have managed to “de-extinct” a previously lost species.
❌ No de-extinction announced, although, a major breakthrough was achieved with the creation of a 99.9% complete genome of the thylacine from a well-preserved 110-year-old skull…
Synthetic life forms (eg from completely synthetic DNA/RNA) will escape from a lab and be found replicating in the wild.
❌Thankfully none (that we know of…)
So there we go… a finger close to the pulse of many trends, but often ahead of the curve in terms of my estimated timing.
Coming up: my predictions for 2025!
Catch you for the first newsletter of the year shortly… so much to catch up on.
Ngā mihi
Ben