Google's Gemini model is delayed
theverge.com
theverge.com
I don't see how that can make the project move faster. I wouldn't be surprised it has the opposite effect actually.
A tiny example I heard: he overheard an employee (who was working on some urgent project he parachuted into) complaining that it was hard working late in this particular office because the closest microkitchen had slim pickings by the end of the day. He stops the conversation and asks a few followup questions. For the rest of the project it was always fully stocked with all the best snacks and drinks, milk and cereal always topped off, fruit always available, etc. I don’t remember the exact details (it’s been a decade since I heard this story) but I remember thinking it went above and beyond what I had available close to my own desk. They thought the dinner spread at the closest cafeteria got better too, but they didn’t know if it was a coincidence or if he made it happen.
I’m sure there’s a ton of the “the boss said something and we’re overreacting” effect. But so much of Google has ossified and slowed that I’m sure it’s also helpful to have a walking sudo command to make the org work for you.
Possibly. But you surely already realized that business is a lot about "showing off", in this case showing off that AI is a matter for decision at the top level.
I do think it's a bit ridiculous (although quite possibly true) that they would be wary what advertisers think. The ad platform is full of terrible AI options already.
They did sink capital into Anthropic recently so perhaps that shuffled things up internally / changed some roadmaps?
If that's the case - Anthropic's models are extremely capable and can be easily launched to the top of the leaderboards if they go for a less belligerent training approach.
these foundation models are trained on trillions of tokens to achieve quality. 200 data annotators with unknown quality issues may be not enough..
See, that's the difference between OPENai and old g
I had it answer me things like "this model, which costs 38k€ is cheaper than this other one, which costs 29k€". I've never seen ChatGPT make such basic mistakes.
Further, booking a flight etc means finding the correct 3rd party aka search.
Likewise with 3rd party extensions; a good LLM will make the best choice based on criteria.
In a way the LLM becomes the end user from a search/advertising perspective. The actual end user just expects an answer, not a list of 5 ads and 10 possible places to go look for the answer.
So yeah, search as an activity still exists, it’s just much less monetizable because the value prop of LLMs is distilling all that noise to a useful answer.
I rarely just want an answer when I am doing a search. But when I do I generally care more about accuracy than I am willing to trust a LLM.
If we’re talking about the viability of the search ads business in the next few months, sure, those are good reasons people will stay with traditional search and therefore be monetizable.
But as an industry trend, I think search will be 100% LLM intermediated pretty quickly. Pics of a subject, projections of hurricane routes… I’d be very surprised if both aren’t provided by LLMs within 12 months.
If anything, I can see this poses more of a risk to websites than to Google. You’re still querying a knowledge graph representation of the internet, it’s only really the last step (clicking a link to a third party website from the SERP) that disappears.
If anything in the next 12 months I expect LLM’s to get worse in many areas as people start doing the equivalent of SEO to get LLM’s to say whatever they want. It’s GPT 5.0+ vs companies trying to sell stuff etc.
That would have taken a while to do manually.
At least until they get blocked.
Here, I just went to chatgpt and asked it two questions -- it now just automatically searches bing:
> can you explain what's currently going on with the israel palestine situation?
ChatGPT The current situation in the Israel-Palestine conflict, as of November 2023, is characterized by ongoing hostilities and a deepening humanitarian crisis. Key aspects include:
Human Impact and Fighting: As of November 13, 2023, the death toll in Gaza has reached 11,100, including more than 8,000 children and women. The collapse of medical centres and communication services in Gaza has made it difficult to update casualty figures. Israeli air raids have continued, and there have been fatalities on both sides, including Israeli soldiers and Palestinian fighters .
Gaza’s Hospitals in Crisis: Gaza’s medical infrastructure is under severe strain. The second largest hospital in Gaza, al-Quds, ceased operations due to a fuel shortage. Al-Shifa Hospital, another major hospital, is struggling with numerous challenges, including handling the bodies of those killed in the bombardment and having to close its doors to new patients. There are ongoing international concerns about firefights in hospitals endangering innocent patients and staff .
Diplomatic Efforts and International Response: Diplomatic efforts are ongoing to address the crisis. The UN has been involved in memorials for colleagues killed in Gaza. U.S. President Joe Biden has engaged in discussions with Qatar's Emir regarding the protection of civilians and the need for humanitarian aid in Gaza. There has been a call for an immediate ceasefire, and Hamas has suspended hostage negotiations in response to how Israel handled the situation at al-Shifa Hospital .
West Bank Developments: There have been reports of fatal incidents in the West Bank, including the shooting of a Palestinian man by Israeli forces. The Palestinian Health Ministry has reported the deaths of at least 186 Palestinians in the West Bank since October 7. Clashes and explosions have been reported in areas like Nablus and Qalqilya .
This situation is complex and continues to evolve, with significant human suffering and widespread international concern. The ongoing conflict, coupled with the humanitarian crisis in Gaza and the West Bank, remains a focal point of international diplomatic efforts and media attention.
User > what started the current fighting ChatGPT The current Israel-Hamas conflict, which escalated into a war in October 2023, began on October 7, 2023. This escalation was initiated by Hamas, the Palestinian Sunni Islamist group, which is designated as a foreign terrorist organization by the United States. Hamas launched an unprecedented and multifaceted assault on Israel from the Gaza Strip, involving attacks by land, sea, and air. This offensive included a massive barrage of rockets fired into southern Israel in the early morning hours of October 7, with the impacts of these attacks being felt as far as Tel Aviv and Beersheba .
These actions marked a significant escalation in the longstanding conflict between Israel and Hamas and led to a severe humanitarian crisis and a high number of casualties. As of mid-November 2023, the conflict had resulted in the deaths of over 12,000 people, including civilians and journalists from various nations .
The complexity and intensity of this conflict highlight the deeply rooted tensions and the ongoing struggle in the region, with each new incident contributing to a cycle of violence and retaliation.
---
It includes links to all it's sources as well (for example -- west bank developments links here: https://www.aljazeera.com/news/2023/11/13/israel-hamas-war-l...)
I don’t see that as success, but I understand why you would argue it was.
Further the assumption that search engines would allow OpenAI to leverage their work while running them out of business is extremely suspect. So if the argument LLM’s will replace search engines while leveraging them is poorly thought out.
If you write such speculative theories, I want to do, too. :-)
What if Google already "37'd" OpenAI, but the result is too dangerous to be released to the public, i.e. it takes a huge amount of work to "defuse"/"cripple" their AI for it to be suitable for "public consumption".
What if Google has an actual GAI that is so good they’ve secretly turned over all strategic decisions to it, and it is directing this “appear to be behind” strategy for reasons known only to itself?
[1] https://en.wikipedia.org/wiki/Person_of_Interest_(TV_series)
> “We have not deployed the system because we think that it has a lot of biases and flaws of other types,” said Irving. “I think the question is, how do you weigh the communication advantages — like communicating with humans — against the disadvantages?
Seems less speculative than a company with more resources, more know-how, more talent, more training data, more compute, more product experience, more anything, somehow unable to ship a competing product. But that seems the common conclusion people draw.
(Clearly, now that I have raised this possibility, particularly weird longtermists must start worshipping Google, just in case.)
if they actually had a superior model they'd release a demo or benchmarks even if they didn't make the model available to the public
We've watched them be total top of the field world wide in machine learning, things they demoed over a decade ago now were basically magic then they proceeded to do absolutely nothing with the tech and just leave it to rot, now look at them scramble to finally try and implement some of that into reality and their organizational structure just isn't up for the task.
Current CEO needs to sit in all ML meetings for a week, figure out who knows how to operate and build it, who's stopping it from leaving the labs, separate the implementation wheat from the bureaucratic chaff, metaphorically lock the wheat in a skunkworks for a month and tell them they have to demo in 3 weeks and ship something when the doors are unlocked at the end of the month.