Richard Munro

@sarbjeetjohal @Microsoft I can't see any mention of reinforcement learning or overall economics impact, but it certainly speaks to the need to focus specific models on specific tasks. My guess is that if it doesn't learn, it will get very expensive over time?

AI News: @Microsoft introduces Critique, a new multi-model deep research system in M365 Copilot. Most interesting part of the announcement: While GPT generates the response, Claude will review the output ⁠for accuracy and quality before presenting it to the user, Microsoft said. The company expects to make that workflow bi-directional in the future, allowing GPT to review Claude's drafts as well. My POV, Microsoft is realizing that they are an OS company with wide distribution of their productivity suite and they can’t bet the farm on any one vendor, especially OpenAI. Did they lose some time, yes — Is it too late for them, no! BTW, while making these changes Microsoft is still trying to save the OpenAI deal/relationship and is keeping the economics of serving AI front center. Microsoft is also launching 'model Council', a feature that will allow users to compare responses from different AI models side-by-side. cc @dvellante @furrier @RikMunro @DavidLinthicum @MarshaCollier @pnashawaty @cleartechtoday

@VMware Definitely worth a read of this independent and vendor agnostic report.

Just in ➡️ Private Cloud Outlook 2025: The Cloud Reset report. See key findings on the increasingly important and strategic role of #PrivateCloud as a platform for both new and existing applications. brcm.tech/4mCP7CE

@dvellante I agree Dave. From September - "allowing customers to bring AI models to where the data resides. In short, Private AI gives you the flexibility to run any AI models adjacent to your data with all your protections in place" :) news.broadcom.com/cloud/how-ai-i…

I predict that the repatriation narrative is going to shift this year to bringing AI to the data. Co's like Dell have a big opportunity here with on-prem AI - data gravity, governance, security

@anthonyspiteri There is a change for sure but couple of thoughts. Maintenance in the real headache along with especially security maintenance. If everyone built their own apps it will be a hackers field day in a years time. Also massive legacy risk - Lotus Notes showed us what can happen there.

I'll expand this out by saying that it's not only customers, but also individuals that can now build their own applications to suit their own needs... be it business or personal. If you have a problem or an idea.. and are able to Prompt Engineer and run basic apps... you are now capable of competing with SaaS.

What if it was actually just a group of smart engineers doing cool stuff?

@AmedeoTritto @cswolf Doesn't affect the downloaded copies, and Chris is talking about PrivateAI. You wouldn't be using the online hosted service.

@cswolf www-cnbc-com.cdn.ampproject.org/c/s/www.cnbc.c…

From the very beginning, our vision for #PrivateAI has been that: 1) AI is moving too fast; you need to invest in a platform that lets you quickly pivot as new models and services emerge. 2) You will get the best TCO by virtualizing and pooling AI capacity, which has never just been in about the GPU (CPU, memory, network, and storage I/O are equally important). 3) Claims that you need hundreds to thousands of GPUs for AI are often not true. We can get 1,000 engineers on 1 GPU for AI code-assist inference. The attention #DeepSeek is receiving today is further proof that investing in an AI platform is your safest bet.

@dvellante @dylan522p In fact, this one says its outside its scope. ChatGPT initially told me everything was perfectly legal.

@dvellante @dylan522p I had an interesting 30 min discussion with ChatGPT over whether using copyrighted data was illegal. It confirmed its responses had been tuned eventually. I realize this is a different level of severity but it's the same base problem.

I don't. If in fact DeepSeek has really made a breakthrough, which I believe it has, it means the denominator for AI just got slashed by orders of magnitude. That is awesome news for productivity. Great news for hyperscalers. Much more concerning for @OpenAI & @AnthropicAI

A moment with momentum! This CIO Dive article quotes numerous tech leaders, analysts, and customers who see the key benefits of private cloud: security, control and resilience. Read more about the resurgence of private cloud. ciodive.com/news/private-c…

@SkarredGhost VR is fantastic to share. They don’t really allow sharing across headsets even in the same household. The software is clunky, pairing for streaming fails a lot with full reset as the fix. Loads worth upgrading to Q3 for - but looks like hardware engineers are running the business

Quest App Downloads Drop 27% Year-Over-Year as Meta Aims to Bolster Quest 3S Sales In my opinion, one of the factors that contributed to it is that many people bought the Quest 2 and these people are not upgrading for now to 3/3S roadtovr.com/meta-quest-sma… #VirtualReality #VR

@sarbjeetjohal @sama This AI thought leader nails it “Anyone can check off answers and put through routine problems. The trick is to expand research and figure out ways to ask the right questions.” Isaac Asimov 1959

Irony of it all, @sama says “Figuring out what questions to ask will be more important than figuring out the answer.” My POV: without having prior knowledge of a domain, one wouldn’t know what are the most important questions which one would need to ask. cnbc.com/2025/01/13/ope…

@JoeBaguley @British_Airways I burned the last of my avios last year. I suspect that sadly they will find out it was actually retaining ‘loyalty’ even with such disappointment in recent years. That’s what BA had left to compete on - it’s not price, it’s not quality, it was loyalty.

Interesting to see that @British_Airways have accepted their inevitable, increasing decline with honesty & open arms. Publicly embracing the fact that they for some time have been treating folks not like passengers or customers but “self-loading cargo”.

@dvellante @sarbjeetjohal Don’t want to be ‘that guy’ but I would at this point cough in a way that sounded vaguely like ‘and what was chatgpt trained on?’ ;) isn’t it ‘fair use’ for generative to create new forms? That’s the defense….

@sarbjeetjohal >>USA funds the innovation for the benefit of Chinese competitors. All this with government support + a willingness to steal IP with impunity = a competitive national force

Alarming: The Chinese Free Model DEEPSEEK V3 reportedly surpasses OpenAI GPT-4o and Meta's Llama 3.1 on crucial benchmarks. This model, trained using Nvidia H800s (less capable than H100), was developed at a cost of only $5.5 million USD in just 2 months. Speculations suggest it might have been trained on ChatGPT output. Watch my video analysis on this development in this post. Link to Video by @dee_bosa is here: youtu.be/NJljq429cGk?si… @dvellante @furrier @Craw @rwang0 @ShellyKramer @rneelmani @RealStrech @SanjMohan @jonfortt @MarshaCollier @theCUBE

@Scobleizer The irony of having to plug weak algorithms by manually doing a task 50,000 times that you wouldn't have had to do without the algorithms, in order to discuss how advanced AI is :) What a world we live in!

I have muted everyone in tech. I am unmuting now that I know that it hurts your reach. But only if you are following me and you leave a comment here. That way I know you haven't muted me. I will leave the rest muted because Elon said that if I bug those who have muted me, I'll be marked as a spammer and penalized. Everyone is on my lists: x.com/scobleizer/lis…

@CTOAdvisor Have a chat with @cswolf - he is doing this day in and day out :) but it’s also just one example. A good one though!

@RikMunro I don't think they fluctuate that much when you are considering inferencing. Re-training or training yes.

Unpopular Opinion: Private Cloud for enterprises is a failed model. Cloud as an operating model is ideal for managed services. It may be the worst model if you host a heterogeneous application set.

@CTOAdvisor AI workloads have very fluctuating demands. Scaling up and down to meet those demands is important. Being able to cost optimize that across requests, and deliver locally to the data sources, and to resource schedule across an array of best fit available capacity.

@RikMunro Why does it require private cloud primatives such as elasticity?

@CTOAdvisor Low latency to data sources, sovereignty of ownership, wide ecosystem of AI tools and models brought to data vs data to tools and full stack preferences. Yes very much. Private AI requires Private Cloud. The cloud operating model, delivered privately. IMO.

@RikMunro It is cloud. Is it a private cloud?

@CTOAdvisor I guess I would posit back that AI is moving at a much faster pace, across multiple vendors, and benefits from self service, advanced automation, intra organizational multi tenancy, and resource sharing. Also from privacy, sovereignty of data ownership, etc - Is that not a cloud?

May be too long for a tweet, but the value of the cloud is to provide a control plane for a set of abstracted resources. Cloud providers innovate at a pace that private cloud platforms can not keep pace, and they serve a much larger spectrum of customers. There's little to no incentive for private cloud instances to maintain that pace. If the abstraction ages 10 years without innovation, is it cloud?