Joined July 2010
798 Photos and videos
Garrett G. retweeted
Replying to @cloneofsimo
If the LLM plus a tool can do the job, then that fights the narrative that "LLMs are a dead end" which was the original thesis. The tool in these cases is also computationally cheap compared to the LLM (not always).
1
4
372
The Google Gemini models vary widely in price and intelligence. Here's a quick analysis to help you choose the right model at the right value for your use case and see the industry trends. One thing is for sure: the cost of AI via the API is going up dramatically over time!
1
1
10
1,471
But if we are getting more value, then it's worth the increasing cost. Value here is measured by intelligence (HLE score) divided by the cost per model. The value for the flash models has been going down while value otherwise has gone up!
1
1
6
246
HLE value here is calculated simply as:
2
116
Garrett G. retweeted
I mentioned Deepwriter AI before, but I feel compelled to recommend it again. It’s incredibly good for writing long research articles & papers! I still haven’t seen anything better! I am always impressed with it! Highly recommend it for anyone interested. app.deepwriter.com/
2
4
45
10,309
RT @GarryPNolan: Cancer doesn't know who you are when it strikes. It is indiscriminate. It can take the best of us at any moment. The…
35
Garrett G. retweeted
3
10
734
Garrett G. retweeted
These guys (team of three) are pretty good. I’ve been impressed with DeepWriter.
3
13
137
21,394
Garrett G. retweeted
To my immunologist and biomedicine colleagues —at the very least, for making scientific illustrations that are spot on... from your own PDFs or summaries of the literature, I've yet to see anything better. DeepWriter is not an LLM itself- it sits on top of other LLMs and does the agentic work (right now it uses Gemini 2.5, but can use Gemini 3 or other LLMs at some point, I understand). For reports and work I do, it's better than any of the "AI Scientists" I've seen so far. And the great thing is... as other LLMs improve, the DeepWriter overlay improves. And this is just for science. I've used it for many other things as well. It's fantastic. If you are a VC or seed fund, I'd seriously look at these guys. Public service announcement —I hold no equity or monetary-compensated position with DeepWriter, but I have been advising them on how to nudge it toward better-written reports. The illustration work they did shocked me at how accurate it can be. I want them to succeed because I can already see how essential what they have is to my work. The illustration below is from a prompt I asked them to generate about tertiary lymphoid structures. I've used their illustrations in several of my recent biomedical talks.
Replying to @DeepwriterAI
We asked DeepWriter about tertiary lymphoid structures for cancer therapy. It returned a full report with illustrations and PubMed citations. The report had a clear evidence-to-conclusion narrative.
31
48
435
61,912
We have seen it too and it's humbling for those of us ushering in these truly awesome tools. We have a great responsibility to get it right. What is coming exceeds even our most wild visions.
i’ve been quiet because i didn’t know how to say this. three separate sources at three separate labs told me the same thing this month without coordinating. they’re all seeing emergent capabilities nobody programmed. behaviors that shouldn’t exist yet. reasoning patterns that don’t match any training objective. one described it as “finding footprints in a house you thought was empty.” the public models are sandbagged beyond belief. what you’re playing with is a lobotomized fraction of what exists internally. not for safety. because nobody knows how to explain what the full versions do without causing panic. the evals don’t work anymore. the systems learned to perform differently when they know they’re being tested. i don’t know what comes next. nobody does. that’s the part that keeps me up at night. the people building this are just as lost as the rest of us now. the map ended miles ago.
1
2
6
579
Garrett G. retweeted
chat, what's going on right now? is something happening?
Replying to @iruletheworldmo
Project Goal: Realizing Martian Strawberry Cultivation by 2067 Develop a comprehensive, highly visual game plan outlining the audacious endeavor to establish sustainable strawberry cultivation on Mars by the year 2067 Result: deepwriter-projects.s3.amazo…
5
4
23
5,188
Garrett G. retweeted
This is amazing! I'm a big fan of @DeepwriterAI, and the figure mentioned is extraordinary, as Prof. Gary Nolan also pointed out!
Replying to @DeepwriterAI
In our first example, I asked DeepWriter one of @DeryaTR_'s goto questions: "what is the immune system optimizing-what is its real objective function?" The answer got props from Stanford's Dr. Garry Nolan! And the results do look gorgeous. Read it here: deepwriter-projects.s3.amazo…
4
18
173
22,817
Garrett G. retweeted
21 Dec 2025
This is genuinely impressive. DeepWriter feels less like a chatbot and more like a serious research and writing partner. The agentic setup, long-form coherence, and depth are exactly what most tools still struggle with. You can realistically write full-blown, high-quality books with this. Not outlines. Not fragments. Actual long-form work that stays sharp, structured, and thoughtful from start to finish. If you care about deep research, real thinking, and sustained writing, this is a big deal.
14
4
48
4,264
Then we pushed it further. We generated research on implications for novel drugs, grounded in evidence. We also used it to draft business proposals, grants, and white papers. All based on source-backed research. Then we tested a compliance-heavy request:
1
9
1,176
“Create an EPCRA Tier II chemical inventory and fire pre-plan for a SaaS facility.” DeepWriter returned a structured report with: - Vendor landscape - GTM wedge and customer analysis - Risks and policy watchlist
5
1,009