Charles Dobson 🇺🇦🇨🇦 (@LitigationKM) header
Charles Dobson 🇺🇦🇨🇦 (@LitigationKM) profile picture

Working on the future of civil litigation. Former litigator, now litigation KM, pricing, and innovation. Amateur coder & fledgling ML practitioner. Views own.

Toronto, Canada
Joined March 2020
Photo by Charles Dobson 🇺🇦🇨🇦 (@LitigationKM)Photo by Charles Dobson 🇺🇦🇨🇦 (@LitigationKM)Photo by Charles Dobson 🇺🇦🇨🇦 (@LitigationKM)Photo by Charles Dobson 🇺🇦🇨🇦 (@LitigationKM)Photo by Charles Dobson 🇺🇦🇨🇦 (@LitigationKM)
Charles Dobson 🇺🇦🇨🇦 retweeted
Across Canada, our justice system is under attack. In British Columbia, the New Democratic Party is erasing the law society, eliminating with it the independence of lawyers.
1
1
2
230
What does this mean for law? 🤔
This quoted post is unavailable.
83
Charles Dobson 🇺🇦🇨🇦 retweeted
Kharkiv. Not Beirut. Happens almost daily. Not making news anymore.
386
3,325
280
9,821
619,626
Charles Dobson 🇺🇦🇨🇦 retweeted
My oldest son is missing. I’d appreciate some shares.
This quoted post is unavailable.
83
2,292
63
1,503
164,561
Charles Dobson 🇺🇦🇨🇦 retweeted
this record I made is so fun!! I need the world to hear it!!!
6
38
8
129
294,206
Charles Dobson 🇺🇦🇨🇦 retweeted
This is a great paper. LLMs cannot plan and reason on their own -- they're pattern-matchers. But that doesn't mean they can't be useful for reasoning. They work best *in conjunction* with systems that can actually plan or reason -- such as symbolic planners.
📢 So our paper on LLM-Modulo Frameworks (arxiv.org/abs/2402.01817) got a spotlight nod at #ICML2024. Expect to see me holding forth in front of our🎗️beribboned🎀 poster on 7/23.. (As for the oral--that will be part of the tutorial icml.cc/virtual/2024/tutoria… the day before.. 😋) [Pardon this not-so-humble-brag. It has been a while since I wrote the full paper, did the rebuttals, and put together the camera ready version.. 🤓]
21
238
8
1,362
131,907
Charles Dobson 🇺🇦🇨🇦 retweeted
Annual Twitter Tradition. Fathers Day and Pride in June. People can have complicated feelings this time of year so I'll make the same offer I make every year. If you like the LAST TWEET of this thread, I'll msg you to say, unironically, that I love you and am proud of you. 1/4
1
2
9
1,535
Charles Dobson 🇺🇦🇨🇦 retweeted
This chart is a great rebuttal of the view that RAG will solve hallucination. In the legal domain, it turns out that it does improve precision somewhat, but at the expense of recall. Really great paper by Stanford & Yale researchers assessing legal AI. dho.stanford.edu/wp-content/…
30
205
38
1,025
454,116
Canada will be among these "small jurisdictions" too.
UK and US law have infected LLMs (according to Linklaters). I shudder to think what LLMs have learnt from being trained on US contracts: - big slabs of verbose text - pointless legalese The broader conclusions are not surprising: LLMs don't produce reliable legal advice.
2
231
Charles Dobson 🇺🇦🇨🇦 retweeted
Do the RAG GenAI systems of the big law tech companies give the “hallucination-free” drafting assistance they claim? @chrmanning & Mirac Suzgun aided a @Stanford RegLab study showing they do much better than GPT-4 but still hallucinate about 1 time in 6. hai.stanford.edu/news/ai-tri…
3
26
4
91
26,648
Charles Dobson 🇺🇦🇨🇦 retweeted
The talk about hallucinations in LLMs has gotten it all wrong. The true hallucinations are by company execs who think it is OK to release to general users products that are based on LLMs that confabulate wildly, as all LLMs do. Time will show a high price paid by society.
23
70
11
283
34,845
Charles Dobson 🇺🇦🇨🇦 retweeted
Think about this as you run your usual shopping errands this weekend. #RussiaIsATerroristState #PutinIsaWarCriminal
⚡⚡⚡ Two dead, at least 11 injured as a result of a Russian shelling of Kharkiv, according to Regional Administration authorities. "Apparently, the target was a shopping center where many people were inside. This is pure terrorism," says Mayor Ihor Terekhov. 📹: Radio Svoboda
2
7
13
842
Charles Dobson 🇺🇦🇨🇦 retweeted
Google, FFS.
95
484
277
4,984
1,729,749
Charles Dobson 🇺🇦🇨🇦 retweeted
For those interested in watching via Zoom, on June 13 at 12 noon, the @ONCA_en will determine whether the public has any right to know how many cases go to trial or how long they take to get there, and whether the open court principle really applies in Ontario.
1
7
14
1,035
Charles Dobson 🇺🇦🇨🇦 retweeted
Below is an explanation of "Agentic" workflows in ~ 15 LOC. At its core, "Agentic" just means LLMs that can call functions. Another ex of how unnecessary jargon confuses people github.com/fastai/lm-hackers…
33
62
6
525
50,340
Charles Dobson 🇺🇦🇨🇦 retweeted
Our paper dissecting AI hype in law by @sayashk, @PeterHndrsn and me is now officially published. journalcrcl.org/crcl/article… Legal Tech is a well established industry and language models offer many avenues for improving existing products. But a robot lawyer isn't on the cards.
9
77
5
246
38,933
We're not yet scratching the surface of what can be done in law with current LLM tech, but interesting to note this potential underlying limit that we may bump up against eventually.
In the late 1960s top airplane speeds were increasing dramatically. People assumed the trend would continue. Pan Am was pre-booking flights to the moon. But it turned out the trend was about to fall off a cliff. I think it's the same thing with AI scaling — it's going to run out; the question is when. I think more likely than not, it already has.
158
Charles Dobson 🇺🇦🇨🇦 retweeted
The first @SCJOntario_en Annual Report in more than 5 years just dropped! And it has stats! ontariocourts.ca/scj/files/a…
2
17
22
4,153
FYI, Ontario law people. LSO has released guidance re generative AI. Futures Committee Report to Convocation, April 25, 2024 (azureedge.net)
1
149
Charles Dobson 🇺🇦🇨🇦 retweeted
that’s right, there is no guarantee of significant improvements, especially since LLM’s are now stealing from each other contaminating the data with their own drivel.
Replying to @GaryMarcus
I kept hearing people saying that "Right now, they're the worst they will ever be" on the assumption that they can only get better. No! Google Search was better than it is now twenty years ago. There is NO guarantee large data models will get better.
9
14
2
68
9,349